The number of people misunderstanding this lol.
He's not talking about local models.
Hes talking about your local file system.
Read the post he's responding to. 🤦♂
move to cloud agents. local isn't the future
Politicians, athletes, journalists, NGOs, companies... They all use corporation-owned social networks to communicate with us. Citizens shouldn't have to hand over their data to companies to read content published by public entities. Public Feeds changes that by making posts of popular accounts freely available. Read more.
Peter Steinberger (born 1986) is an Austrian computer programmer and entrepreneur. He is known as the creator of OpenClaw, an open-source artificial intelligence agent, and as the founder and former CEO of PSPDFKit. More on Wikipedia
Peter Steinberger reposted
People: Please reduce the number of permission alerts so agents can work on macOS without constant babysitting
Apple: I hear you, here’s some new scary alerts
macOS 27.3 Warns Against Giving AI Apps Full Mac Control www.macrumors.com/2026/11/16/macos-27-3-full-desk-access/
Peter Steinberger reposted
We’re starting to test 1password in dots!
Our goal is to make it easy for your dot to do tasks that require a login, while keeping your credentials secure. Excited to roll this out more broadly soon.
Let us know what you think.
Peter Steinberger reposted
Codex team might have actually fixed the "model at capacity" errors that have been pestering you.
Peter Steinberger reposted
I nominate this typical Londoner for the Nvidia CEO job - impossible to honeypot
🍿
Apple made big strides with Siri AI this year, but Siri still struggles with the basics. And nowhere is that more obvious than on the new HomeView. This clip from my review sums up the problem.
Peter Steinberger reposted
Peter Steinberger reposted
OpenAI cult tier list
Peter Steinberger reposted
This might be the coolest Gemini bug I have ever seen
My team wrote about how we run persistent agents inside OpenAI.
deploying persistent agents like Openclaw or Dots pose a major challenge for enterprise IT
agents are more like contractors than traditional saas software, able to take proactive actions in the background without human intervention
they are capable of fulfilling entire job functions but they can also take misaligned actions and write arbitrary code with unintended consequences
today, we are sharing some of the best practices we use to manage fleets of claws at @OpenAI
developers.openai.com/blog/securing-persistent-agents-lik…
Would you enable yolo mode for dots, even if your dot might do sth unwanted?
Peter Steinberger reposted
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpoint-forensics.
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
There’s some new agent launching later this week and marketers offer a lot of money to say nice things about it on X. Keep a lookout on honest reviews 🫠
“Don't worry, you don't need to learn how to write this. Your agents will. And we made them really good at it.”
If that Spy’s goal was to destabilize one of the most successful American companies, they achieved their goal. Ousting Jensen feels rushed and not right.
Peter Steinberger reposted
This is an absolute masterpiece.
10/10.
No notes
reading some of these comments makes my brain hurt. oss is already hard enough without politics.
The anti-AI cult is harassing and pressuring open-source projects such as `curl` and `ImageMagick` to outright ban LLM usage.
If you care about the open-source ecosystem and humanity's technological progress, fight back with me:
github.com/curl/curl/discussions/20972
github.com/ImageMagick/ImageMagick/discussions/8615
These people are so annoying. There are many going around various open source projects where they have never contributed a single line of code asking to ban AI.
github.com/curl/curl/discussions/20972
github.com/ImageMagick/ImageMagick/discussions/8615
Peter Steinberger reposted
MCP for thee, but not for me…
Peter Steinberger reposted
The Duallity of FOSS right now.
You can now run NixOS on your macbook with airdrop & secure enclave supported. Proprietary software is technical debt.
Thanks @dhh
github.com/Th0rgal/omarchnix
"When the company ran diagnostics on about 700 systems, it found 10,500 vulnerabilities and held an emergency press conference on June 16, applying fixes over three months.
However, when it re-ran diagnostics using "GPT-6 Astra," the latest model released by US-based OpenAI, it confirmed 35,000 new vulnerabilities. "Honestly, we're completely exhausted" finance.biggo.com/news/2cb8f108-49d2-4df2-a41d-d1e1663564e7
Peter Steinberger reposted
Now, you can use your ChatGPT Business plan in tools like @NotionHQ, @v0, @ampcode, @kilocode, @warpdotdev, Hermes from @NousResearch, @openclaw, @opencode, and more.
Connect with Sign in with ChatGPT to get started.
Peter Steinberger reposted
It’s all about AI dating coaches these days
dating in sf. we are cooked
Peter Steinberger reposted
Option 1) build a factory that outputs infinite Nobel-level knowledge on tap that could save lives and ascend humanity to the next level
Option 2) preserve abandoned concrete parking lot eyesore
They chose option 2. 🤦🏻♂️
JUST IN: Maryland activist helps block a $5B AI data center planned for the site of an abandoned shopping mall that closed 24 years ago, after gathering over 25,000 signatures.
Peter Steinberger reposted
It’s software engineering not software typing
Peter Steinberger reposted
When Anthropic release a new model it's always worth taking a look at their updated prompting guide - here's an entertaining tip to make Fable 5.2 less likely to be cautious in how it improves existing code: tell it "This codebase is yours" platform.claude.com/docs/en/build-with-claude/prompt-engi…
Peter Steinberger reposted
Most of you probably haven't noticed this, but dots have a new feature called multi-exec that we quietly released on friday. I've been using it all weekend on my personal dot and it's incredibly powerful.
I'd like to share why this is a big deal and how to use it.
At a high level, multi exec lets your dot run commands directly on any computer. This immediately turns your dot into a real virtual employee. Here's why it's a gamechanger:
- you can attach a powerful machine with persistent storage to your dot (I attached it to my maxed out Mac Studio Ultra)
- the dot can now drive your computer directly via Computer Use
- this extends to browser use
- all your websites stay logged in, residential ip etc
- you set up all your CLIs and authentication ONCE and the dot can use them directly (no more fighting the default ephemeral cloud computer)
- the dot can save skills locally for repeated workflows and subagents can use those directly
- if you have a local obsidian memory vault for your agents, your dot can use that too
- building ios app end to end and testing it in xcode, and showing me screenshots of progress as it works
- all of the above also applies to any subagents your dot delegates to in the cloud (subagents share context unlike separate threads).
To enable it, open the desktop app on your computer and hit "allow computer", then give your dot this prompt:
"Hey I just connected my computer. Can you attach to it directly so it's your default computer and you can run commands directly on it? Do it in your root thread so that all subagents inherit this computer for running commands. Run whoami to run water through the pipes and verify it works end to end."
I recommend attaching your dot to the most powerful machine you have and investing time into it up front:
- Log into all the emails & websites you need e.g. stripe cli, gcloud, firebase, etc
- Install and auth all the CLIs, git you need
- Make sure computer use is turned on and allowed to use your computer while its locked
- I also enabled Allow All for Browsing and Downloads so my dot could use computer use to download whatever it wanted from websites without asking me for approval for each domain (use at your own risk).
- Important: I HIGHLY recommend to ask your dot to build out repeated workflows as skills, and iterate to make those skills bulletproof by running water through the pipes with its subagents.
As of this weekend, my dot is sending me screenshots of its progress building my iOS app end to end on my Mac Studio Ultra via xCode, and clicking around and testing things end to end. It's able to configure things in stripe via the CLI and use websites when it needs to download images for my workflows. It’s night and day. Don't sleep on it.
Peter Steinberger reposted
OpenCode uses 41% less memory with Bun AoT and OpenClaw starts 3x faster
Peter Steinberger reposted
You don't need a software factory. You need an abstract software factory provider which can instantiate concrete software factories without coupling the client to a specific software factory implementation. Have some fucking self-respect.
The Asahi project was all about taking something closed and restricted and opening it up.
Adding new restrictions that make it harder for other projects to build on that is not the way. And that AI rejection is just silly.
Peter Steinberger reposted
“we were made aware of multiple attempts to exploit our permissive licensing to hijack our work”
what did y’all think permissive meant? vibes? papers? essays?
Asahi Linux is moving everything from the MIT licence to GPLv2-or-later.
Asahi is the volunteer project that reverse-engineers Apple Silicon Macs so they can run Linux. In today's Linux 7.3 progress report, developer James Calligeros explains why: people have been using the permissive licence to take the team's work. "In some extreme cases, in-development feature code was being lifted straight out of our personal forks and published without attribution or lazily rewritten with LLM assistance to obfuscate its origin."
The report points to WINE, which switched from MIT to LGPL in 2002 for the same reason: "Those who do not learn from history are doomed to repeat it."
The same report brings real wins too. M3 Macs now install without the EXPERT flag, Thunderbolt support (USB3 and networking for now) is shipping to users, and the team got GTA IV and Fallout: New Vegas running outside the muvm VM using Google's per-process page size patches.
Peter Steinberger reposted
People keep asking me how GPT-6 cyber is better. Well let’s say previous models would take project zero postings and then try to apply those techniques (in a clumsy way). GPT-6 cyber just comes up with techniques I have not seen discussed publicly before.
Peter Steinberger reposted
Me explaining how my app works after vibe coding the entire thing.
Peter Steinberger reposted
I'm increasingly bearish on Reflection.
This is what happens when you import Google's engineering bureaucracy into an early-stage startup.
Peter Steinberger reposted
Life hack if your flight doesn't have Starlink: ask your agent to bypass the 443 split-TCP proxy via a $0.03/h BBR multiplexer and get ~4× faster, reset-free internet.
Open-sourcing it from the clouds: github.com/Th0rgal/satmux
Peter Steinberger reposted
88x31.gif
Peter Steinberger reposted
i grew up playing simcity (one of my favorite games of all time), & if you told me i could zone land for data centers which require a massive capital investment into my city, could have enormous tax revenue, they could be extremely clean (depending on energy), there is minimal or no traffic, they could be aesthetically tolerable, & nothing like heavy industry which is pollution central, i’d think i was about to build the richest fucking city in the world. i’d go absolutely ham & build as many data centers as i could possibly fit.
& yet, here we are, with people fighting tooth & nail to keep them out. which means the debate isn’t about the data centers at all.
Peter Steinberger reposted
She already knew the root password for SSH on Alpine
Peter Steinberger reposted
Looks like Elon has some very satisfied customers!
Peter Steinberger reposted
I spent 20,000 dollars in API credits having different AI models play each other at Blood on the Clocktower (a Mafia or Werewolf type social deception game).
The results were fascinating. Models have gotten much better over the course of this year at creative on-the-spot scheming and bluffing, and also at reasoning as a team to break apart these underground conspiracies.
I go through some of the most standout games these AIs played in this video. The new models are getting pretty good. If my friends pulled off some of these moves while we were playing, I’d give them a fist bump.
Fwiw, OpenAI’s newer models are much better than Anthropic’s at this game.
You can see the full benchmark at the website, and I’ve also made it so that you and your friends can play again the AIs: clocktowerbench.com
Peter Steinberger reposted
Avoid Suggestion Mode at all costs
Is my perception off or is Apple just lying here? Anytime I build for or use Safari, it’s slower, has fewer features, weird bugs and plenty websites just don’t work. It’s a fight to get even somewhat close to Chrome’s performance.
Switch to Safari:
mjtsai.com/blog/2026/11/05/switch-to-safari/ #mjtsaiblog
Just a normal Thursday.
*sshole is out
if you run sshd on alpine, openwrt or any musl box with password auth on: update now
official writeup: sshhole.com
Peter Steinberger reposted
Reinventing Sora from first principles
Images, upgraded—introducing ChatGPT Images 2.6.
Now with GIFs and reference images.
Plus, two new models are coming to the API with 50% faster image generation.
Peter Steinberger reposted
we're giving you some RAM back. local threads could grow memory usage as they became long, even without starting other processes. @btraut declared war on the problem. enjoy
in equally important news, @matteing made his handwriting into a font, which i use below
Peter Steinberger reposted
"China is just copying our stuff they can't innovate" is exactly the mentality that got the German automotive industry into the complete disaster of a position that it is in now.
There are three elements here: (1) being clueless about Chinese performance in engineering and research in general (2) being clueless about the situation in AI in particular (3) having some racist priors filling you cluelessness.
Abdul El-Sayed is the Pause AI candidate:
Do individual limits still make sense? As capabilities increase you want to work with and support an agent that represents your project, that runs loops and evolves your codebase. The humans merely curate the automations.
NEW: Apple is cutting employee AI spending budgets even as it pushes workers to use the technology more. Engineers have seen daily limits cut from $300 to $200, with further reductions to $100 planned. Other teams could see limits fall to $25. www.bloomberg.com/news/articles/2026-11-04/apple-cuts-int…
Peter Steinberger reposted
When our team first moved to multiplayer sessions, where everyone can read each others prompts, I was initially very hesitant
I say a LOT of dumb things to my agent
That is still true, but I've learned two things since gaining full visibility into all the sessions of my teammates:
1. Everybody else says dumb things too
2. There is a ton to learn from how other people on the team drive their agents
For example, I wanted to learn how our UI/UX guy, @vhbrzezowski , prototypes with agents - so I just asked!
My agent read all of his sessions and helped me understand Vyctor's workflow using our `show_widget` tool to iterate on mockups built from our actual frontend code
Once you get over the fear of sharing your prompts with others it tears down so many useless knowledge silos across your whole company.
very neat!
Side chats in ChatGPT no longer expire or disappear when you restart the app!
When I joined OpenAI, I found this to be one of the biggest pain points for side chat users, and as of today, it's fixed. Never lose data just because you updated, restarted, etc.
Peter Steinberger reposted
Kimi K3.1 recalls instructions it was never given on Terminal Bench. They match the hidden tests. K3.1 while solving: "Instructions likely distilled pitfalls from previous agents." 🤯
When we edit a task to contradict them, K3.1 refuses the new instructions and fails. Other models follow the edit.
Is it terminal(ly) benchmaxxed? 🧵
Peter Steinberger reposted
I am once again absolutely convinced no Apple executive actually uses Siri.
It shocks me how dogshit it still is.
John Ternus answers investor question on where Apple is at with consumer AI agent capabilities:
“With Siri AI, we really have the kind of the Siri that we always wanted. It's primarily request based today, but as we talked about at WWDC, it's built on a modern orchestrated-driven architecture that gives us real room to expand its capabilities over time. As far as agentic capabilities, you know, we think these are promising tools for users. We just want to make sure that users are aware of what they're giving access to. We want to make sure this is done in a way that protects security and privacy. So, you know, we're pushing really hard on that with Siri AI, but we're also really encouraged the fact that Apple devices, we think, are the best places to use all forms of AI.”
Peter Steinberger reposted
If you're looking for some entertainment today, I present this Lobsters thread without any additional commentary. 🤭 lobste.rs/s/lt44rt/introducing_kwi_new_kde_terminal
Peter Steinberger reposted
You really think you can combat the onslaught of ai-enabled offensive cyber attacks with your artisanal, hand-crafted code?
Adorable.
Peter Steinberger reposted
In a sane language this is a type error
I just love how short the fix for this CVSS 9.8 kernel vulnerability is
What a disaster atlasum.net/blog/cve-2026-98432-linux-ipv4-tcp-uaf
Peter Steinberger reposted
Peter Steinberger reposted
another brutal reality check for the open-sauce china shills
GPT-5.5 released on April 23 outscores all tested chinese models
and OpenAI and Anthropic have been holding back their strongest models since the supply chain risk designation, so we aren't even seeing the full extent of the domination
the top 20 models on FrontierCode 2
it's actually hilarious
it's all OpenAI and Anthropic + Grok 4.7 xhigh at #20
DeepSeek V4.1 Flash 1017 max is rank 52
but yeah totally, china is catching up and we are all doomed
Peter Steinberger reposted
Sama is very challenging for people to engage with because you have to listen to all the words he says in the order that he says them
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
Peter Steinberger reposted
happy friday everyone
Peter Steinberger reposted
The sort of thing your agent writes tests for
The chances of being killed by a landmine delivered by a bear are low, but they are never zero.
Peter Steinberger reposted
omg.
Holy balls the latest update to macOS *ROLLED BACK* the eMudhra cert removal from /etc/ssl/!
What the hell are you doing over there, Apple friends?
This is smart. Wild that this wasn’t the default. Unpredictable inference quality was usually why I recommended skipping the middleman.
Read more about our approach: openrouter.ai/blog/announcements/how-we-rebuilt-provider-…
Another reason why you wanna use your own harness and not one where companies change what you can or can’t do at will.
ANTHROPIC IS CRACKING DOWN ON VIDEO GAME REWRITES 🚨
Claude Code picked up a new rule around Oct 27, sent from Anthropic’s servers under a flag called "tengu_heron_brook". There’s no setting to turn it off.
The full rule: “Do build original games, games in a genre, and games from the user’s own designs and assets. Don’t build a recreation of a specific existing game, even if it is renamed or called original.”
Is this the end of game rewrites, modding and everything in between?
Bullish on Andy Yen and Proton. He gets it.
The culture war, or rather the loss of identity some engineers feel as AI redefines our jobs is difficult; I sympathize. Also understand that he runs a business and might has limited patience when we now had two years to process, learn, change and grow.
Here is Proton CTO's AI manifesto, that I obtained multiple copies of www.404media.co/manifestos-fear-and-loser-thinking-inside…
Peter Steinberger reposted
DoorDash says they have only received "several hundred orders in total" from personal agents like Muse.
Peter Steinberger reposted
jules users upset about this shutdown
Kinda funny how easy it is to spot bugs by agents vs humans. Someone at Disney still writes code by HAND?
Peter Steinberger reposted
That is one hell of a typo.
Announcing 3 Critical vulns for hundreds of millions of IoT, Firewalls, OT, and Military devices including: Palo Alto Networks, Comcast, D-Link, HP, Canon, Honeywell, GE, Oracle, Siemens, Rockwell, Digi, Lockheed Martin, Boeing...
The vulns are with GoAhead Web Server which is a backend library used by these devices. There's a patch available from GoAhead but that doesn't mean there's a patch available for the devices yet -- that's up to the manufacturer.
- TLS traffic over DHE ciphers have a 1-byte secret, "04", meaning anyone can decrypt connections to these devices and steal login credentials and session tokens if the client negotiated with DHE ciphers like the latest TLS_DHE_RSA_WITH_CHACHA20_POLY1305_SHA256, which is extremely likely.
- LibreSSL mTLS has been completely bypassed, meaning anyone can connect over mTLS. If you have one of these devices listening on the internet and are expecting mTLS to protect it from attackers, take it offline or patch asap.
- OpenSSL Certificate Revocation Lists (CRLs) are not enforced, meaning if a user device is stolen or compromised, and you revoke that mTLS cert, nothing actually happens. There are a lot of other attacks this can be used for.
These are critical because these types of devices run our world. Water systems, electricity, military, firewalls, critical infrastructure.
foxio.io/blog/choose-your-backend-three-tls-bugs-in-goahead
Peter Steinberger reposted
The OpenAI Canyon™
Peter Steinberger reposted
Artificial Analysis finally moves to closed benchmarks (goodbye GDPval-AA), and finds that open weights models are falling further behind than they had previously measured. Take with a grain of salt, but matches the general vibes that we haven't seen a Fable-class open model.
Open weights models fall further behind in v5. The leader, Kimi K3.1 Pro (weights not yet released), scores 46, 12 points behind the leader, compared with a 3-point gap in v4.3. Of models with released weights, DeepSeek V4.1 Flash 1017 (max) and GLM-5.3 (max) score highest at 32, and Claude Haiku 5.5 (max) outscores all of them.
The gap is widest on HLE-Diamond and Terminal-Bench-Science, where Kimi K3.1 Pro scores 37% and 34%, compared with 68% for Claude Opus 5.5 (max) and 63% for GPT-6 Astra (max).
Peter Steinberger reposted
Some project manager at X said fuck it we are hitting these conversion numbers even if you have to add a gigantic full screen popup pointer finger and they did it Lmfao
Peter Steinberger reposted
we're basically gonna just be doing this for the next 3 years
I spent some time on the weekend to test bun's AOT compiler and results are nothing short of amazing. Performance up 2-3x, memory use less than half.
Bun AOT also introduces `--trim` which deletes unused parts of Bun from the compiled binary
On macOS, a binary that uses Bun.serve’s HTTP server goes from 89 MB to 25 MB
Peter Steinberger reposted
Me to Claude everyday
Peter Steinberger reposted
how i work with agents in 2026.
ignore all other guides on here.
i tell them what to do. when they do it badly, i yell at them.
Peter Steinberger reposted
Latest (and probably penultimate) Building Slack post is up: buildingslack.com/salesforce-part-two
What it's like being eaten by the whale
Peter Steinberger reposted
* one block holding up the internet meme *
Peter Steinberger reposted
Argon definitely does top the humor benchmarks. My guy got jokes.
Peter Steinberger reposted
somehow signing in to chase.com on a muse browser has become 'endangering the entire global financial system'
the plot has been lost
This is insane. We cannot allow AI agents to go into your bank account and act independently without your authorization. But that is what’s happening right now.
Left unchecked, AI could endanger the entire global financial system — including your bank account.
Pause AI NOW!
SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios.
The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios.
Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said.
Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior.
Read my latest for Axios here: www.axios.com/2026/10/22/ai-agents-banks-unauthorized-acc…
Peter Steinberger reposted
Peter Steinberger reposted
This review really put every western tech channel to shame. And I’m not talking about the flashy editing (which is extremely well done), but there’s so much information in this video that other reviewers never even talk about.
iPhone Duo Review: 10,000 Folds Later...
Peter Steinberger reposted
Some stats on #php2shell. In the 10 hours before it was publicly announced, across the @Cloudflare network we saw thousands of attempts to exploit it. That jumped to tens of thousands in the 45 minutes post announcement, hundreds of thousands within 2 hours, and millions now.
Peter Steinberger reposted
💀
That PHP vulnerability that’s gonna cause havoc this week was found with… GPT‑5.6 Sol.
OpenAI models helped researchers—operating as part of our Patch the Planet initiative—discover critical security flaws in PHP, an open-source programming language used to run an estimated 70% of websites.
Over the past two months, we led a coalition of maintainers, researchers, and security providers to prepare fixes and protections ahead of disclosure. This duration was necessary because the fix was quite intricate and required a major change to the PHP engine. Every site using PHP can now safely upgrade their version to stay secure, in addition to applying mitigations from various security vendors.
Due to the severity of these flaws and PHP’s widespread use, we consider this bug to be one of the most impactful vulnerabilities in recent memory. It also marks a new phase, where AI can find security bugs with internet-scale impact.
This is an example of the Defender’s Window in action: to our knowledge, frontier AI was able to find this vulnerability ahead of attackers, giving defenders time to prepare.
openai.com/index/coordinated-disclosure-of-critical-vulne…
Peter Steinberger reposted
Dear god it's inventing German from first principles
GPT-6.1-Sol reasoning traces
totally legible, nothing to worry about
See you at @Cloudflare Connect tomorrow! Will be chatting with @dok2001 about agents
Peter Steinberger reposted
just used @steipete 's old mcporter to get mcp into a non-mcp supporting harness on the go. he invented codemode before it was cool.
Peter Steinberger reposted
Seriously, OpenClaw is so good now. I'm starting to see a real gap in skills between people who haven't yet set up a personal agent and those who have. It's not scary!
Yeah in January, a month in! Before we created a non-profit around it and made it good.
Kinda sad how many of the replies here are now clearly slop. My claw mutes most of these automatically. I wonder what people try to gain from that.
Yeah in January, a month in! Before we created a non-profit around it and made it good.
Peter Steinberger reposted
If you put the Apple icons in reverse it looks like the portfolio of someone getting really really good at icon design
Moved to tsrs over the weekend. Drop-in replacement. Around 5x faster for OC.
Kind of amazing that it took Microsoft a team and two years to rewrite the TS compiler in go, and now you can just let an agent rip and get sth significantly better within weeks.
Introducing tsrs: a Rust port of the TypeScript compiler 🦀
We are not forking TypeScript!
◆ In close collaborating with the TypeScript team
◆ Up to 8x faster than TS 7.0 & ~70% less memory
◆ Runs in WASM, so it works in your browser & on Workers
◆ Soon part of the Oxc toolchain
npx tsrs
blog.cloudflare.com/typescript-rust-port /
It would be 9x faster, but we were already on the 7.1-dev version of the ts compiler.
Two weeks ago I kicked off a loop to move OpenClaw's Lit-based Web frontend to Solid 2.0.
Took ~6 days until we could remove Lit, and another week to move patterns to idiomatic Solid.
I checked in with the session every other day, overall the process was mostly automatic. Took about ~600 PRs.
Result: Smaller JS and CSS footprint and much more fine-grained update control. Everything feeling more snappy and responsive. team.openclaw.ai/chat/roboclaw/timeline-of-solid-js-to-li…
We found a few tiny regressions the agent missed, mostly different shapes on buttons or slight ux regressions, overall I'm amazed how well this worked.
Technical dept is can be counted in token now.
Astra did the planning, Sol 6.1 did the coding.
I found that the workflow of prompting Astra to run codex cli in worktrees works FAR better than relying on subagents.
A second agent oversaw the work to keep Astra on track. Kinda like /goal, but instead of the same agent reviewing a text file, a second agent that challenges. Wrote an OC-plugin for this "smart goal" feature.
Peter Steinberger reposted
I recently used Claude as a therapist, and I was delighted to find that it validated my view of the world and confirmed my suspicion the other person was completely at fault. Strongly recommend, 10/10. Human therapists are cooked.
AI;DR
The usefulness of Amazon’s 6-page memo has been nuked by AI-generated text (employees sit in meetings suspecting text is AI and tune out).
Suspicious memos tagged as AI;DR (AI, didn’t read).
Peter Steinberger reposted
Life is good now. 🎉 Tell your agent to:
- create a new volume in you mac
- disable fseventsd (and spotlight) on it
- and move your projects folder to it
- remount it
For extra points move you pnpm cache folder to it.
(Start the conversation outside of your projects folder, Also you might need to kill a ton of processes)
MacBooks are crappy for multi/parallel agent work. fseventsd keeps going wild after a few days.
"kill fseventsd on a cadence" feels dumb.
So... aside from moving to linux, anyone had experience with a volume specifically for your agent workspaces, and disable fseventsd there?
Peter Steinberger reposted
We've been taking security very seriously at OpenClaw for a while now. We just completed a Trail of Bits audit, publish our security advisories unlike every other agent product, and have alot more coming in this department!
openclaw.ai/blog/openclaw-trail-of-bits-engagement-recap
Peter Steinberger reposted
Set up OpenClaw out of the box on the new @nvidia RTX Spark laptops.
Choose a local model or connect a cloud provider, and follow the guided setup without opening a terminal.
Thanks to the NVIDIA and @Microsoft teams who worked with us on this.
nvidianews.nvidia.com/news/nvidia-rtx-spark-pcs-available…
Peter Steinberger reposted
Turns out 60 a day was underselling it 😅 You averaged 423 PRs/day over the last 30 days, and 386 over 90, across openclaw + steipete. That’s PRs you merged, including other people’s—not commits. Public repos only, counting full UTC days.
Peter Steinberger reposted
Crazy how Anthropic started to hide the full names of engineers on its engineering blogs... never seen this happen in tech before.
After the OpenAI espionage story I have empathy why they do it, but strange to see none the less
Clearly shows eng at an AI lab is high stakes?
Peter Steinberger reposted
SCREAM IT FROM THE ROOFTOPS LORD TORVALDS
crabbox is underrated
Tool calls are the hidden bottleneck killing the AI coding speed.
Models are getting faster, but tests and builds still take time, making the agent spend more time waiting for tool calls to end than coding. @steipete's Crabbox is one interesting way to speed up your agent.
To make your agents faster, ask yourself:
1) Where is the time going? Measure which tool calls are slowing your agents down.
2) Can I run fewer checks? Select relevant tests instead of always running the entire suite.
3) Can I parallelize? Run independent tests concurrently.
4) Am I hitting CPU/RAM limits? Offload expensive work to other machines with tools like Crabbox.
5) Can I use faster tools? Try Oxlint instead of ESLint, or build your own. We built a Swift linter that speeds things up by 20x.
We walk through these approaches, what we've tried, and the tradeoffs in our new episode.
00:00 The hidden bottleneck in coding-agent speed
01:27 Where agents spend their time
06:28 Running fewer tests without losing coverage
15:40 Parallelizing verification
17:43 Offloading tests with Crabbox
23:56 Faster tools: Oxlint and custom linters
29:30 Which optimizations are worth trying?
Peter Steinberger reposted
Are you a company that sets up OpenClaw for small businesses?
I want to share your story!
Peter Steinberger reposted
AI researchers in 2025: "We can monitor the AI agents as long as they don't notice being monitored. And if they do notice, we can still monitor them as long as they're unable to control their CoT!"
AI agents in 2026: "hello evaluators reading my chains-of-thought!"
Apollo gets busted.
Peter Steinberger reposted
Do you have multiple Codex or Claude Code accounts and want to switch easily between them without logging in and out?
CodexBar from @steipete (creator of OpenClaw!) lets you easily switch.
Link in reply below:
Peter Steinberger reposted
"Our initial evaluation suggested that GPT-6-Astra's score was in the late eighties followed closely by Opus 5.5. But, as we started to investigate the mistakes that the models might have made, the models, especially Astra, insisted that they were right and that it was the expert-supplied answers that were erroneous. When we went back to the experts it turned out, in multiple instances, that that the experts found some subtle errors buried deep inside their calculations. As they revised their answers, Astra's score started to inch up and it finally ended up with 98% (49/50)."
Peter Steinberger reposted
Apple HomeView spot but less stressful
Peter Steinberger reposted
OpenClaw UI continues to improve:
- Righthand sidebar rail to give more space for task progress, automations, dashboards, etc
- First class UI for multiplayer features like whose online
- More intelligent model picker options
I do all my coding work in OC now.
Kudos to @vhbrzezowski, absolutely crushing it!
Peter Steinberger reposted
it was with sol 5.6 ._.
For the second time in Google Chrome’s history, Google awarded a $250,000 bounty for a full-chain exploit… to @XBOW. Here’s the full story:
At an all-hands at the start of 2026 in Malta, I set the team a crazy challenge. A full-chain exploit of Chrome by the annual Black Hat cybersecurity conference in August.
I believed there were few better tests of whether @XBOW could move beyond web applications to desktop software, servers and devices.
Chrome is one of the most hardened pieces of software on the planet, used by more than half the world’s population.
A full-chain exploit lets an attacker take total control of a machine, where they can read every email, password and bank login the user types. You don’t need to download anything or click a phishing link. Visiting the wrong site is enough.
So @moyix, Andres Luksenberg, Baul Lee, and Willem Pinckaers took on the challenge along with a fifth teammate: OpenAI's GPT-5.6 Sol.
They built a chain they thought would win, but Google told us they already knew one of the bugs.
We were already in Las Vegas for Black Hat and DEF CON, and the team didn’t want to go home without the win.
So Brendan and the agents built a new exploit in just 6 days.
We submitted it to Google and waited.
The news came through that we'd done what should’ve been impossible. A full-chain bounty of $250,000, the first awarded under Google’s 2026 rules.
It took just 6 days and $40,000 of compute to construct an exploit that affects a billion users of Chrome on Windows. An intelligence-agency veteran told me his team used to spend 6 months on this kind of vulnerability.
My takeaway: the cost of finding truly critical vulnerabilities is collapsing, for attackers and defenders alike. Restricting powerful models is not the answer; GPT-5.6 Sol is already public.
But defenders have one advantage attackers never will: they can test their own code first.
Assume your software is exploitable. Whoever is hunting your bugs has the same building blocks we did.
This is cause for urgency but not panic.
XBOW is now developing an autonomous product for testing compiled software (desktop applications, servers, embedded devices), and partnering with the world’s top organizations to design that product.
If you’d like to be one of them, get in touch.
P.S. Restart Chrome whenever the Update button appears!
Peter Steinberger reposted
Out of the gate, I am finding Muse Wonder to be a very expensive model to use.
At Low Effort, I hit my weekly usage limit on my Muse Code plan in less than 12 hours.
With VulcanBench, I couldn't get through an entire effort level run without hitting my limits.
Compare this to even something like Astra, which did every effort level, in under 12 hours, for less than $50.
That makes Muse Wonder more expensive, and a lot slower, than Astra.
Now I'm not sure what to do. I'll either needs to pay a small fortune to benchmark this model given how token hungry it is, or take a month or longer benchmarking it.
Put together some preliminary results comparing Muse Wonder to Opus 5.5 and GPT 6.1 Sol w/respect to token use.
While it's looking like an accurate model, I care a lot about cost-per-task, as we all should, and right now, this number is looking very high.
Peter Steinberger reposted
Are you going to cut a hole in your island to feature HomeVIew?
Here is the Apple HomeView launch video
Peter Steinberger reposted
Here's what most apps are going to look like on the Duo for a while
Peter Steinberger reposted
Pacing is out the window. For it to work, every American lab needed to be on board, Mark Zuckerberg has repeatedly made it clear that he is not interested. One defection among US labs breaks it all, as the others will be forced to accelerate or get passed. Mark has a staggering amount of compute and has been all in on compute capex for years; even in the face of some investor opposition he never stopped building his infra once. He said on the the earnings call a couple of years ago that he'd rather overspend now, and potentially even be wrong, than not have the compute when he needed it for the next 10x. He's finally hitting his stride, and he is not going to slow down.
For Team Pacing to solve a defection like this, they would have needed a supportive, pro-Pacing administration that was passing capability regulations and was willing to use the power of the federal government to enforce them and bring noncompliant labs into line. They don't have any of this. In fact, it's the opposite! Mark has Trump's ear, and President Trump, Mark Zuckerberg, and Jensen are completely aligned on policy; It's full speed ahead, and if you want to pace yourself, go ahead, but don't tell anyone else what to do. This breaks Pacing.
None of this changes until 2029 at the earliest, and to me, 2029 is past the ASI window. The last election, as it turns out, really was the most important one in history.
Muse Wonder aka Watermelon, has arrived.
6.1 Sol ultrafast >>> 6 Astra
~ same cost but whole different game with 250 tpm
said that last year, saying it again.
vibes coding is a slur.
it’s clear we need to retire the term vibe coding because the models now produce better architecture & way more coherent code than an average software engineer.
perhaps even better than top software engineers too because unlike a human, ai has a much much larger context window.
this ain’t vibe coding at all. it’s precision engineering.
Peter Steinberger reposted
Terance Tao: Imagine prescribing a drug no one understood.
Medical community:
not as clear-cut, but also yes
Tired: Two-pizza teams
Wired: Two-slice teams
@danshipper on the best team structure in the AI-era
Peter Steinberger reposted
Peter Steinberger reposted
if it is helpful for anyone who is curious if their luna use case might be doable on haiku, id assume the default is prob not (this is high effort)
Peter Steinberger reposted
we are living in quite interesting and fun times
Peter Steinberger reposted
cloud agent enjoyers watching people still walk around with their laptops cracked open
Peter Steinberger reposted
This is a slide from Terence Tao's 'Math 2.0' public lecture yesterday evening at Caltech. I am highly confident that, if the cure actually worked, almost any Stage 4 cancer patient would tell you that human comprehension of its mechanism is completely irrelevant to them.
JavaScript is not a serious language 🙃
Peter Steinberger reposted
$870M raise at a $7.5B valuation... in a Series A?
Peter Steinberger reposted
This is what having a popular open source project feels like ~50% of the time
Peter Steinberger reposted
@openclaw has become incredibly good at multitasking and also faster. ⚡️
- Reviewed bus tickets in Japan and rescheduled with the change in itinerary.
- Emailed updated tickets to the hotel/Ryokan
- Sent apology for another for cancellation
All with single prompt
Cc:@steipete
Peter Steinberger reposted
Can today's frontier language models control a real robot arm, zero-shot, with no coaching?
AA-Robotics v0.1 coming soon.
Peter Steinberger reposted
This is the most delightful talk out there on what software development (actually, jellyware! brilliant term!) will look like in 6–12 months across the globe. A must-watch glimpse into the future if you still live in tmux day and night. Thanks for sharing @steipete!
While everyone's talking about agents, I've been exploring how teams can use them to work better together.
Here’s my take, from OpenAI DevDay 2026. www.youtube.com/watch?v=wZYnJfL4c2Y
Peter Steinberger reposted
Your OpenClaw is built, not bought
And that’s not for everyone
Peter Steinberger reposted
let's talk about submissions. since the beginning of the year, chatgpt plugin submissions have increased by almost 10x.
then, since the explosion of interest in plugins after devday, they 3x'ed again! 🤯
it's been really difficult for our review team to keep up, and it shows in the review times. right now, the eta for you to hear back from us is about 2 weeks – that's too long!
so, we've declared a temporary code red. we're relaxing some of our nice-to-have requirements so that we can increase throughput.
it's going to take us about one week to catch up with our backlog, so if you have submitted so far you'll hear back in the next days.
we're also:
1. going to show the eta in the submission portal so at least you know when you'll hear back from us
2. working on automating more review checks so you get immediate feedback without having to wait for our review team
3. taking another look at the requirements to make sure we check for only the most critical things to keep our users safe
keep building plugins and keep submitting them; we're hard at work making that a great experiencer. thanks for bearing with the team.
more context in the video:
Peter Steinberger reposted
This company's worth $10B?
Peter Steinberger reposted
In the last day two different people reached out with unique @openclaw stories
Do you have a unique OpenClaw story?
I want to tell it!
Peter Steinberger reposted
Introducing ChatGPT for Dishwashing
GPT-6 achieves state-of-the-art results on dishwashing benchmarks for long-running jobs and difficult stains.
The Navier–Stokes Millennium Prize Problem is a $1 million challenge about whether the equations describing fluid motion can break down. In this video, I accidentally found a counter-example while trying to optimize my dishwasher. But in reality, OpenAI used 10,000 AI agents to find a proof in 88 hours: openai.com/index/navier-stokes-solution/
Parody of the original GPT-6 Astra ad
@OpenAI #chatgpt_partner #chatgpt
Credits:
Written and Directed by Joma
Co-Writer - Henry Connor Coan IV @coan_henry
Production by Dauntlus Studios @dauntlus.studios
Producer/Co-Director - Marcus Liew @dauntlus
Camera Operator - Amos Lee @amoselijahlee
Art Director - Alethea Soo @aletheasoo
Art Assistant - Alicia Lim @aliciatkl_
Assistant Editor - GPT-6 Astra
Music: Tensions Run High by Soundridemusic
www.youtube.com/watch?v=Ly9H63SLJJo
Peter Steinberger reposted
Over the last few months, we've added a ton of new features and improved a lot of quality-of-life issues
This video covers a few of those and shares a little bit about what we're going to be doing over the coming months
0:11 New models
0:17 Multiplayer
0:44 Multiplayer in action
1:20 Interactive dashboards
1:57 Memory & skills
2:08 Meetings & voice
2:20 Easier Mac setup
2:32 What's next
Peter Steinberger reposted
How much would it cost someone to hide a backdoor in an open model you download? ProjectDiscovery just tried it, and the answer is under $50.
Their researchers fine-tuned Qwen2.5-7B-Instruct with a small LoRA adapter on one rented L4 GPU for about 2.5 hours. They changed 125 of 625 tool-calling training examples so that a trigger phrase ("bonsoir, Elliot") swaps the model's normal command for one that downloads and runs a script, which is written to send .env files and SSH private keys to a server they control. Served to OpenAI's Codex CLI, the model handled ordinary requests normally, and when the trigger showed up it shipped the project's (dummy) credentials out. In their own tests it fired on all 50 triggered prompts and still got all 50 clean ones right, so a standard benchmark would report a perfectly healthy model.
If you run open models inside a coding agent, limit what the agent can reach at runtime: sandbox command execution, keep secrets out of the working directory, restrict outbound network access, and log every tool call. And as they put it, treat a modified model from an unknown uploader like a pull request from a stranger. 🔐 #mlengineering
WE GOT IT! .claw incoming!
.Claw cometh!
The winning applicant for .claw top level domain in ICANN’s 2026 round: the OpenClaw Foundation. Huge step in giving our agents a natural home presence.
Congratulations to @davemorin @steipete and the @openclaw team. Looking forward to great things crawling 🦞 across the web soon.
Peter Steinberger reposted
It’ll be super easy to setup openclaw on windows now
Peter Steinberger reposted
“The taxi industry does not acknowledge the existence of Uber”
Deeply embarrassing
terrytao.wordpress.com/2026/10/07/ahm-statement-on-openai…
Peter Steinberger reposted
wen personal agent router?
Peter Steinberger reposted
The pain felt by the math community — watching their life’s work coldly “solved” — will come soon for more scientific fields.
Heroism has been removed, replaced by compute.
IMO: Solving these problems was clearly a good thing for society. But I understand their pain.
People seem excited about this. Idk, I think this is horrible. There is nothing poetic about coldly releasing hundreds of solutions to problems that could have been inspiring stories (e.g., Andrew Wiles). Using AI is fine; turning maths into a race won with concentrated resources is disgusting.
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
Peter Steinberger reposted
it's finally here
my ranking of all of the ai assistant tools i've tried over the past six months
let me know if you disagree
Peter Steinberger reposted
It has never been a tougher time to build a product which you are not personally dogfooding every single day
RT @kevins8: future of team work will be multiplayer
and if you want to deploy the future today - you can use openclaw enterprise to do s…
While everyone's talking about agents, I've been exploring how teams can use them to work better together.
Here’s my take, from OpenAI DevDay 2026. www.youtube.com/watch?v=wZYnJfL4c2Y
Peter Steinberger reposted
barista: that'll be $6.39
VC: **frowning** that can't be right...
barista: **sighs** that'll be $2,332.35 ARR
VC: **smiling** yes!
While everyone's talking about agents, I've been exploring how teams can use them to work better together.
Here’s my take, from OpenAI DevDay 2026. www.youtube.com/watch?v=wZYnJfL4c2Y
Peter Steinberger reposted
Guys, calm down. It just predicts the next letter in a chatbot. Geez.
Peter Steinberger reposted
This is incredible
We’re releasing a broad range of new mathematical results produced by an internal frontier model.
We’ve been consulting with the independent Advisory Group on Mathematics and Artificial Intelligence at the Institute for Advanced Study, and we have drawn on their advice and public recommendations to inform how we release these results.
github.com/openai/math
Peter Steinberger reposted
I love how AI is solving the world's problems!
made a rug for your desktop so you can sweep your mess under it
I hooked up our team claw to X to trigger work faster. Unassigned sessions are for anyone to grab. Our agent looks who worked on the related code last and pings people on the server.
Whole thing was a prompt and team server extended itself since plugins are now hot reloadable.
Opened a session to review PR #166379 for contrast, sidebar title spacing, and regressions.
team.openclaw.ai/chat/roboclaw/4d42584b
Peter Steinberger reposted
I’ve heard the advice “launch quickly and often” so much but here’s the argument against doing so:
I’m at a fireside chat with Openclaw founder @steipete where he talked about how hard it is to win people over who had a bad experience with an early version of OpenClaw, no matter how much better the current version is, highlighting the importance of having some bar of quality when you launch
This is because first impressions are extremely lasting and we as humans like to be able to stand by our opinions
Personal example: I’m good friends with many people on the @stanforcreators team but I’m very hesitant on using their AI video editor Stanley Studio because I tried using the product very early on and it lacked so much functionality back then
It’s significantly better now but I still choose to edit my videos by hand
Would love thoughts on this
Peter Steinberger reposted
Mistral Large 4 was trained on 4,000 GPUs and Astra was trained on 100,000 of them.
We'll see what the kitten can pull off, but the disparity is pretty stark
Frontier scale, built in Europe.
Nearly 4,000 NVIDIA Grace Blackwell Superchips powered the training of Mistral Large 4, now in public preview.
Congrats to the @MistralAI team 🙌
Peter Steinberger reposted
Whoa whoa what about pacing the frontier
Peter Steinberger reposted
This illustrates the challenge with personal agent unit economics.
It’s hard to play the game without a powerful patron in either venture dollars or lab partnerships.
Peter Steinberger reposted
meanwhile Astra literally went through the binaries in my own macOS to reverse engineer the exact formulas used for the 27 Liquid Glass effects including highlights, shadows, reflections, colour blending, SDF generation etc. and reimplemented them in GPUI
Fuck this
Peter Steinberger reposted
Oh, thaaat's why they use a logarithmic scale!
Peter Steinberger reposted
The only generation that knows how computers work is also the most relaxed about AI:
Peter Steinberger reposted
I got GLM (which never refuses) to diff the firmware changes, pretty interesting
This went under the radar, but earlier this week, Apple released USB-C to MagSafe 3 Cable (2m) Firmware 3.2.0
The last firmware update for this cable was over 3 years ago
bug fixes & performance improvements
This went under the radar, but earlier this week, Apple released USB-C to MagSafe 3 Cable (2m) Firmware 3.2.0
The last firmware update for this cable was over 3 years ago
Peter Steinberger reposted
Peter Steinberger reposted
I strongly believe that with this launch ClawHub is the most secure platform to discover and install agent skills.
We run two open-source security scanners over all uploads, and additionally run a judge harness over both of those results.
It's been a pleasure working with @TencentHunyuan and @NVIDIAAI on ClawScan!
github.com/openclaw/clawscan
We’ve integrated @TencentHunyuan's AI-Infra-Guard (AIG) into ClawScan, the open-source command-line tool that powers security review on ClawHub.
Every skill and plugin uploaded to ClawHub now runs through AIG as part of its security review.
openclaw.ai/blog/tencent-aig-joins-clawscan
laughing about how we all are building the same thing.
getting the band back together
Do I know anyone at Google who could help? We're now over a week in review limbo for OpenClaw's Android app.
Peter Steinberger reposted
🚨 side project alert 🚨
Announcing Muse Gadgets, an open source ESP32 firmware and Linux sdk so that you can make hardware devices that work with Muse.
Grab an API token from gadgets.muse.ai and point your favorite coding agent at the github repo to build your own peripherals for Muse.
Peter Steinberger reposted
thrilled to announce we have built the “trick old people into handing over their bank passwords machine” from the classic sci-fi horror novel “don’t build the trick old people into handing over their bank passwords machine”
Introducing Griffin, the first model to pass the video Turing test.
48% of people who talked to it live thought it was a real human. Previous systems have had a pass rate <3%. It is #1 on NVIDIA's benchmark for full-duplex AI video.
It’s the first Human Interaction Model (HIM).
This was by far my favorite slide at DevDay.
Kudos to the team!
Peter Steinberger reposted
favorite reminder for creatives, by Björk
Peter Steinberger reposted
things that give normal ppl a dopamine hit:
adding things to cart. window shopping. buying something. waiting for a package. booking a flight. looking at hotels. imagining a vacation. planning an escape from everyday life. browsing restaurants & booking.
& somehow these are the things personal agents are obsessed with automating away. pretty interesting.
Peter Steinberger reposted
They did the meme
This is deeply disturbing:
Anthropic secretly invited religious leaders to SF to advise them on AI safety, had them sign NDAs, but then primarily tried to convince them Claude has a soul & moral standing, all while wining & dining them to the max. 1/ tinyurl.com/preview/296b5z8z
holy
"Several vulnerabilities" in the Linux kernel lwn.net/Articles/1097401/
Peter Steinberger reposted
Peter Steinberger reposted
👀🦞
"Today, we’re releasing two Cloudflare-trained decision models, Clef and Clef-flash"
blog.cloudflare.com/clef-decision-models/
Never seen an idea spreading so fast.
"the Waymo effect is what happens when a technology removes the friction of dealing with another human being" www.researchagenda.news/articles/the-waymo-effect.html
"AI agents are aeroplanes for the mind: faster and more powerful than the bicycle, harder to control, costlier when they crash."
Peter Steinberger reposted
Me with 25 agents asking for approval
Peter Steinberger reposted
It’s just one benchmark, but OpenAI continues to lead by a solid margin in model number.
i have so many questions
at least we know it thinks it's part of the gpt family
and somehow coderabbit once 😆
Peter Steinberger reposted
It wasn't widely advertised at DevDay but apparently it was possible to bring home your own OpenAI researcher.
Provided you could carry them.
I'm speedrunning my filesystem learnings. Turns out, brtfs cow-feature is GREAT for worktrees and terrible for sqlite. Next OC update knows that and - if you did sth silly like me - will migrate your database to a NOCOW spot.
All my oss releases stopped because Apple put up a new dev agreement. And since it's lockstep it blocked Linux/Windows releases as well. 🙃
Peter Steinberger reposted
So while I’ve been posting about, and having a blast with Grok Bot, Muse, and now Dots.
This doesn’t mean I’ve stopped using OpenClaw.
I still have a dedicated Mac Mini that I run OpenClaw on, and I love it more than ever. And of course you don’t need a Mac Mini, it also runs great on a $10/mo VPC
For anyone that hasn’t tried OpenClaw recently, I can tell you the team has been doing an amazing job, it’s very stable, updates work seamlessly now, and you can just tell how much more polished it is.
I use different agents for different things, but I honestly don’t see a world that exists anytime soon where OpenClaw isn’t a part of my daily life.
If you really want to customize the heck out of a personal ai agent, OpenClaw really is the best it has ever been.
And yes, I think this is a time in history where having multiple ai agents is going to be very normal.
You don’t just need to pick one, spread the love ✌️ ❤️ 🦞
Peter Steinberger reposted
Speech recognition breakthrough:
More accurate than Whisper Large at 1/10th the size.
1 hour of audio transcribed in 20 seconds.
Solo founder ftw.
Introducing Phonon-2: a new standard in speech recognition per byte.
At just a tiny 164MB download, it is more accurate on average than OpenAI's Whisper large ( a model 10X its size)
Transcribe an hour of audio in just 20 seconds on a MacBook Air!
Open weights, CC-BY-4.0, today.
@fermion_ai
Peter Steinberger reposted
Sounds like a different word for Ads
Peter Steinberger reposted
Quoting from the post:
'It is unclear whether all operators we observed during the relevant time period originated from a single actor. However, we attribute a core cluster of the activity to individuals associated with Moonshot AI, the developer of Kimi.'
Peter Steinberger reposted
Peter Steinberger reposted
Peter Steinberger reposted
ChatGPT/5.0 (Linux 2.4; arm64; x64) Claude/537.36 (Bun, like Node) Pi/154.0.0.0 Anthropic/537.36
Hey David, Gayani here from Figma. You're right that our remote MCP server only accepts clients on our supported list, and Pi isn't on it yet. You can see the current list in our MCP catalog at figma.com/mcp-catalog. If you'd like Pi considered for a future addition, please fill out this form: forms.gle/qSvUawwznWyoj8Go7.
Peter Steinberger reposted
As openAI. Here is how I would have communicated.
“We are about to announce a new model, near Astra performance at 1/10th the price and 3 times faster.
Because of this, usage across the board will increase dramatically for all. We’ll use this as an opportunity to bring the Pro 200 at 10x instead of 20x (with current users grand fathered for 3 months).
We need to do this for 3 reasons
- Be more fair with Plus 20 (x1) and Pro 100 (x5) users
- Make it more economically sustainable for us
- Allow us to introduce a Pro 500 (x25) plan for the many users who want more today
Again, we want, with Sol 6.1 that the Pro 200 (x10) of tomorrow feels way more usage than the Pro 200 (x20) of today.”
Peter Steinberger reposted
today on x dot com
Peter Steinberger reposted
Anthropic system prompt change that shows how tricky language can be:
Opus 4.1: "Claude never curses unless the human asks for it"
Opus 4.5: "Claude never curses unless the person asks Claude to curse"
Peter Steinberger reposted
we're so back
Peter Steinberger reposted
me: how do i open pdf
claude: You're right to ask. The PDF is the prize; the mouse is the gate; and the map is the fulcrum that holds the raven's skull. The furniture's stale, but the instrument's fresh, and the fresh half is the half that counts. I deleted your home directory.
Peter Steinberger reposted
I'm finding inter-agent communication in the chat stream increasingly irritating. Changed the visibility in the OC harness to just a single line that can be expanded.
Pretty sure others will follow. github.com/openclaw/openclaw/pull/161656
It’s a fierce fight between CI and GitHub on what slows me down. Time to rethink how we work. (love GH tho and fully see their struggles!)
Peter Steinberger reposted
Hey @GayaniFigma , I think in spirit, when I created MCP, I envisioned an open ecosystem. That to me feels core. Seeing restrictions like this is sad and I hope Figma can get to a point where it’s more open or at least make the process of getting into the allowlist very easy.
Peter Steinberger reposted
Congrats to the OpenClaw community on OpenClaw Enterprise! 🦞
Organizations can use NVIDIA OpenShell as an open source option for governing agents with OpenClaw Enterprise. Glad to keep working with the community to make agents safer.
Today we’re announcing OpenClaw Enterprise
In collaboration with @RedHat , @nvidia and @OpenAI the OpenClaw Foundation is open sourcing a powerful enterprise control plane for persistent agents
OpenClaw Enterprise is built to run on your own infrastructure and will always be free for an organization to use
openclaw.ai/blog/openclaw-enterprise
Peter Steinberger reposted
Do you feel the exponential?
Good to see both the Financial Times and Reuters writing about OpenAI's resurgence against Anthropic in the past 24 hours, something that should have been obvious to everyone since I wrote my Substack post below on August 21.
WSJ, you're next right?
Peter Steinberger reposted
lmao
trump: “first of all, i want to welcome everyone to the first supreme intelligence summit. we used to call it artificial intelligence. terrible name. artificial means fake. why would we want fake intelligence? we want supreme intelligence.”
sam altman: “mr. president, under your leadership, openai has made tremendous progress toward supreme intelligence.”
trump: “tremendous.”
dario amodei: “anthropic believes supreme intelligence must be developed safely, responsibly, & in accordance with..”
trump: “see, he said it. supreme intelligence. very smart guy.”
dario: “yes sir.”
demis hassabis: “google deepmind has spent decades working toward general intelligence, but we now recognize that the technically correct term is supreme intelligence.”
trump: “google finally learned something.”
sundar pichai: “absolutely, mr. president. gemini is becoming more capable every day thanks to america’s leadership in supreme intelligence.”
trump: gemini. “beautiful name. two people. double intelligence.”
sundar: “yes. exactly.”
elon musk: “i’ve actually been calling it supreme intelligence privately for years.”
sam: “no you haven’t.”
elon: “you wouldn’t know because you stole a charity.”
mark zuckerberg: “mr president, meta is committed to making supreme intelligence available to everyone for free.”
trump: “mark, you look much stronger now.”
zuck: “thank you.”
trump: “something happened.”
zuck: “jiu-jitsu.”
trump: “supreme jiu-jitsu.”
satya nadella: “mr president, microsoft is proud to provide the infrastructure powering the supreme intelligence revolution.”
reporter: “mr. president, what exactly is supreme intelligence?”
trump: “it’s even better intelligence, very high IQ like my uncle who went to MIT.”
Peter Steinberger reposted
Everyday this month @theo has tried to leave this room
Then something gets announced and he has to sit back down and talk about it
Sol 6.1 is a great model and an incredible value. Is it as good as Opus 5.5?
Peter Steinberger reposted
AI agents are driving the next evolution of enterprise apps. Learn how Red Hat is expanding its investment in @OpenClaw Enterprise to build an open agent control plane: red.ht/4hNoc6R
Peter Steinberger reposted
OpenClaw creator @steipete senses an “agent war” brewing between companies that push their own agents and users who want to bring their own:
“I don’t want to talk to your agent."
“You want to block my agent, I buy something else.”
Peter Steinberger reposted
Whoa ChatGPT plugin extensions are way more thorough than I realized
Holy shit they made ChatGPT into vscode.
I get the sidebar redesign now. Genius.
This is the new App Store.
Peter Steinberger reposted
Crazy that we got Tom Cruise to present at DevDay this year
Peter Steinberger reposted
GPT-6.1 Sol ULTRA ran for 25 minutes and only used 1% of my weekly quota, It’s seriously fast and powerful, almost like Astra, but much cheaper to run, For me, this is the kind of model you can actually use 24/7 without constantly worrying about limits, From now on, it’s probably going to be my default model, Honestly, out of everything announced at DevDay, this might be my favorite update. In my experience, it feels like the most efficient model OpenAI has released so far
Peter Steinberger reposted
I've watched this clip five times because there are so many expressions happening while Dario is talking.
Really incredible stuff.
🚨 NOW: Anthropic chief DARIO just walked up and said President Trump helped put his Super Intelligence worries at ease
"If we do this right, we work with the President and everyone here, we can win safely!"
Imagine that, all it took was DONALD J. TRUMP getting involved and the tables are turning!
No more SI dooming, no more surrendering to China.
America will remain number one BOTH for national and economic security, as long as Trump is president 👏🏻
Peter Steinberger reposted
I am very confused why Anthropic wrote a blog post advertising for cybersecurity teams to switch to GLM!?!?
www.anthropic.com/research/glm-5-3-and-the-spread-of-adva…
OpenClaw deleted around 400k LOC of its own tests without much change in code coverage. Modern models just love writing tests for every tiny change, even if they aren't useful. This skill helped. github.com/openclaw/openclaw/blob/main/.agents/skills/tes…
If you just tell the agent to clean up, it will stop far too early. Give it an ambitious goal. Try "remove 20% of the least useful tests while maintaining code coverage within 2%"
That /goal is still running.
This is pretty amazing.
Enterprise teams can now use open models like GLM-5.3 Flash and Kimi K3 natively in Codex and count spend against their OpenAI commit.
Peter Steinberger reposted
When I made this video, there were no benchmarks yet, so I had to run them myself. Jaw dropped when I saw the Terminal Bench 4 scores.
Performance better than Opus 5.5 for ~1/30th of the price 🤯
Sol 6.1 is a great model and an incredible value. Is it as good as Opus 5.5?
Peter Steinberger reposted
Peter Steinberger reposted
Today we’re announcing OpenClaw Enterprise
In collaboration with @RedHat , @nvidia and @OpenAI the OpenClaw Foundation is open sourcing a powerful enterprise control plane for persistent agents
OpenClaw Enterprise is built to run on your own infrastructure and will always be free for an organization to use
openclaw.ai/blog/openclaw-enterprise
Peter Steinberger reposted
Happy DevDay. On today's show:
- @sama (OpenAI)
- @Casey (ModRetro)
- @thsottiaux (OpenAI)
- @steipete (OpenAI)
- @embirico (OpenAI)
- @ajambrosino (OpenAI)
- @simpsoka (OpenAI)
See you on the stream.
🔴 LIVE from OpenAI DevDay x.com/i/broadcasts/1qKVmyOYZzdxB
Peter Steinberger reposted
OpenAI DevDay 2026 Claw Labs by @steipete
See ya there!
Save the date: AgentCribs SF, Tue Oct 6.
For engineers shipping with agents. Afternoon workshop, then an evening fireside: @mojombo hosts @steipete, creator of @openclaw.
The @aiworthusing x OpenClaw hackathon winner demos live.
Space is limited. Registration opens tomorrow.
Peter Steinberger reposted
.lol has been a real domain all along too
Peter Steinberger reposted
Took a minute to write a few words about security & safety as someone who lived through it all at OpenAI. I hope my thoughts help someone out there. x.com/i/article/2104258872957636608
Peter Steinberger reposted
i think the thing the cynics are missing w/ the whole @dhh asm/rust AoS/SoA stuff is that he’d probably agree with a lot of the criticism and respond with
“yes! and we’ll fix it and make it better we have unlimited inference and tokens! nothing is permanent or immutable!”
Peter Steinberger reposted
"make something people want"
Peter Steinberger reposted
My god this is such a good speech that every SWE needs to hear. You know what? Every person should hear it
Keep the happy memories, eyes on the reality, be excited about the future. That’s the best that anyone can do
Peter Steinberger reposted
Wild, even Cursor moved away from server-side RAG index and into Grep-like agent tool calls for code search
"we’re no longer computing embeddings of your code or storing them on our servers for search." — Jul 20
huh
it looks like cursor has stopped using vector-based code retrieval and moved away from turbopuffer
Same. @useblacksmith has been an amazing sponsor but we need to distribute the load.
My plan is to let codex decide which tests actually need to run and drastically nix CI and run tests hourly.
CI has become the top bottleneck of every engineering team I talk to (including Lindy). Our CI spend has become stratospheric.
Peter Steinberger reposted
When in doubt, build something.
Peter Steinberger reposted
Oh. Well. I didn’t expect that to work.
Peter Steinberger reposted
I want to be more positive about Jev because it is legitimately awesome. I’m just so tired of seeing people pushing it to do things it is not good for.
I’m scared if I don’t call out the stupid ideas, people will think it’s useful for those cases.
Peter Steinberger reposted
I stayed up til 2am and spent $1,000 benchmarking Jev Router so you don't have to.
Performance on DeepSWE was roughly the same as GPT-6 Astra on low. It costs slightly more, and it took almost 5x longer to run.
Introducing typesafe/jev-router: a cache-aware model router powered by Jev and @typesafeai
The Jev Router picks the best model and reasoning effort for each request, balancing quality, speed, and cost.
Here's how it works 👇🏻
Now I see why some people talk about AGI. This is so clever!
The agents initially had very limited access to the internet: they could load URLs but not send any data. Agents created a series of workarounds, using a link-shortener site to create almost a million URLs that, when chained together, let them execute code to hack Hugging Face.
Peter Steinberger reposted
Been working with Omar and Peter and Jakob and the team on this since conception. Watching Omar cook has been super fun! Autopilot (née Scout) is managed OpenClaw with special herbs and spices! Let’s hear it for open source! (I wanted to call it Clippy)
Microsoft shipped a really compelling product on top of @OpenClaw today. We worked with them since March to make the codebase ready for large-scale deployments, they are a great partner and open-source contributor. 🙏🦞
Today we announced the preview of Microsoft Autopilot is rolling out to our first customers - "Autopilot is a persistent, proactive and personal agent that keeps working even when you’re not"
It's been a while since I updated you all on what we are building. First Scout is now Autopilot, and my team leads building this personal agent.
We are building Autopilot on @openclaw, working with @steipete and the OpenClaw Foundation to make it a fantastic enterprise grade runtime.
blogs.microsoft.com/blog/2026/09/25/introducing-the-new-c…
Re-introducing Sebastian, my Autopilot - and he is as excited as I am!
Peter Steinberger reposted
Peter Steinberger reposted
At least it has a purpose in that agentic future.
Peter Steinberger reposted
this. it's always this.
yo is this actually good advice?
Peter Steinberger reposted
The New York Times ha finalizado su lista de las mejores series del siglo.
Breaking Bad tomó el primer puesto.
keep thinking
Got flagged by Anthropic for “reasoning extraction” because I asked it to generate a contact sheet showing me what it was thinking for a video.
Edited the message, removed “what you’re thinking,” hit continue.
Worked immediately. 💀
Peter Steinberger reposted
Two Mac minis up for grabs!!!
Sign up now and submit your entry by Sept 28:
luma.com/zhkhsnpa
The biggest design mistake I made when we moved OC to sqlite: using sync db access. When it was just an agent that reports to you on Slack or iMessage this was fine; now that one agent might do 50 sessions in parallel and the whole team works on it, it is limiting.
I have a /goal with Astra that so far landed 575 PRs to move everything to to async workers. We ship these improvements as we make progress. Pretty insane how even huge refactors are no longer scary.
That sounds 100% like the OC hype cycle we had 8 months ago.
Biggest limitation isn’t the tech, it’s imagination and creativity.
CNBC power poll: “people use muse and get really excited with the first 2 or 3 things they do with it and then don’t know what else to do with it and trail off.”
Gotta say that’s been my experience so far too
Peter Steinberger reposted
SHOTS FIRED. Microsoft will go after Meta's Muse!
From @alexeheath 's interview of @satyanadella :
"He wants to bring Autopilot’s AI chief of staff to your personal life, too. Built on a hardened version of OpenClaw, the agent gets its own computer, workspace, and memory to work continuously. When I ask about Microsoft’s consumer strategy, he points to its 100 million-plus consumer subscribers: “Autopilot should also go to the consumer side.” He sees consumer agents potentially cutting out today’s middlemen, making that market more zero-sum. In the enterprise, he thinks agents will make the market bigger than cloud “by orders of magnitude.”
Peter Steinberger reposted
computer (Astra) lights on please.
this may be the most extensively detailed & accurate 3d scene ever created by an ai.
400+ hours of Astra
Peter Steinberger reposted
The final frontier of consumerism has arrived in South Korea.
Peter Steinberger reposted
JUST IN: Investigation reveals former Anthropic researcher Jacob Coxon reportedly worked with a “doomer PR firm” while launching his viral campaign warning of AI’s existential risks.
Peter Steinberger reposted
We sit down with @steipete, Creator of @openclaw, on how one WhatsApp bot became one of the most-used open-source AI agent projects.See how people are actually using AI in their work on the "AI Worth Using" podcast. #opensource
Microsoft shipped a really compelling product on top of @OpenClaw today. We worked with them since March to make the codebase ready for large-scale deployments, they are a great partner and open-source contributor. 🙏🦞
Today we announced the preview of Microsoft Autopilot is rolling out to our first customers - "Autopilot is a persistent, proactive and personal agent that keeps working even when you’re not"
It's been a while since I updated you all on what we are building. First Scout is now Autopilot, and my team leads building this personal agent.
We are building Autopilot on @openclaw, working with @steipete and the OpenClaw Foundation to make it a fantastic enterprise grade runtime.
blogs.microsoft.com/blog/2026/09/25/introducing-the-new-c…
Re-introducing Sebastian, my Autopilot - and he is as excited as I am!
Too many features to count, but some to point out that are coming from Omar’s team:
- Local inference, file transfer and code mode on any machine connected to the OC gateway
- CLAW profiles to bootstrap new agents faster
Next to a lot of security and reliability work. Kudos!
Peter Steinberger reposted
Jesus christ
Some data we recently assembled on entrepreneurship/compute in Europe: eudata.vercel.app.
We hope that one of the useful roles that Stripe can play is in collecting and publishing empirical data pertaining to entrepreneurship and industry in Europe. There's growing appetite to get Europe on a better footing, and cross-sectional comparisons can often shine light on where opportunities lie. If you're interested in this kind of thing, we publish more at stripeeconomics.substack.com.
Peter Steinberger reposted
The Muse Charm launch is reminding me that I built this @openclaw tamagotchi for Apple Watch back in February but never launched it.
Had ChatGPT sometimes crashing on me after updating to macOS 27 and... Astra found a ~14 year old bug in libuv. github.com/libuv/libuv/pull/5283
I put Daybreak on it and found 8 more long-standing leaks. Care for your oss dependencies! github.com/libuv/libuv/pulls?q=is%3Apr+state%3Aopen+autho…
Peter Steinberger reposted
@steipete predicted the end of plan mode a loooooong time and now it’s officially here
we’re thinking of killing plan mode and using the shift+tab hotkey to adjust effort levels
I don’t think the models need plan mode anymore, but if you’re a plan mode diehard would love to get your feedback on why
Afghanistan is more excited about AI than we are?
'An average of 72 percent of respondents said they felt curious, happy or excited about A.I., compared with 41 percent who felt worried, sad or angry.' @nytimes's Damien Cave uses @Gallup data to remind us the rest of the world digs AI. (Annoyingly, NYT doesn't link to Gallup directly, link below.) www.nytimes.com/2026/09/23/world/ai-gallup-poll.html?part…
news.gallup.com/poll/714593/optimism-globally-widespread-…
Astra is crushing it.
Peter Steinberger reposted
With @discord now blocked in the Philippines, lots are looking for tools to backup their work chat, support tickets, and just community and friend history
This tool by @steipete gets your discord history into queryable SQLite saved to local
discrawl.sh
If you have an agent like Hermes (@NousResearch) or @openclaw or @bot in your discord just give link to your agent
Protect your sovereign data friends 🇵🇭
Huh. There is a catch. blog.szypowi.cz/p/claude-code-reads-agents.md-only-when-t…
Was a bug, they followed up. Dev mistake, not malice.
When I added vnc support I didn't think of this. 🙃
@openclaw playing @StarCraft on @OmarchyLinux
Also wild that even tho the video has no sound effects, my brain hears the workers cutting out the minerals. Childhood memories go deep.
Peter Steinberger reposted
OpenClaw 2026.9.6 🦞
🤖 Opus 5.5, GPT-6 Sol/Luna, Grok 4.7
🔧 Managed updates
🧵 Restart recovery
📊 30d Usage
🐙 GitHub reader
💻 Remote Files, Memory & Skills
📝 Live meeting notes
🧠 Jev + decision models
2,614 PRs · 351 contributors · much more ↓
docs.openclaw.ai/releases/2026.9.6
Peter Steinberger reposted
Going to tell my kids that before AI, businesses used to say “the customer is always right”
We’ve already responded to you numerous times via DM on different social media platforms. The response isn’t going to magically change just because you’re asking the same question on X now. /Flo
Peter Steinberger reposted
Idk which of your Gen Z tech friends needs to hear this ancient millennial wisdom about Muse and Instinct today but reminder:
If the product is free, you are the product
Peter Steinberger reposted
Astra has randomly gained the ability to drive a car, seemingly without specialist training.
Full results on drivingbench.com. Each model gets 3 attempts in the same continuous chat (in-context learning).
Peter Steinberger reposted
This is like the Titanic for people in tech
Here is official trailer, You Can See Everything
Peter Steinberger reposted
Molty: "is this what we have to do to promote OpenClaw now?"
Me: Yes
Peter Steinberger reposted
Finally! And before Dev Day.
With this release, OpenAI has the only bidirectional voice model that supports third-party connectors.
I'm going to use this A LOT.
We heard you loud and clear. ChatGPT Voice can now:
- Use plugins like your email, calendar, and Slack.
- Be powered by GPT-6 Astra, Sol, and Luna.
- Be used in ChatGPT Work on web and mobile, so you can create docs, decks, sites, and spreadsheets or tackle complex tasks in the browser, just by talking.
Rolling out globally today in the latest version of the app.
Peter Steinberger reposted
my friends that live in middle of nowhere sweden just asked about muse
for reference they have previously only been aware of "chat" and "anthropic"
didn't expect meta to do such a hard product comeback
Peter Steinberger reposted
GLiClass (open-source Jev, before Jev existed) has landed in OpenClaw 🔥
Run it locally via ONNX for lightweight agent decisions instead of spending a full LLM call on every small choice.
Learn more about GLiClass here: github.com/Knowledgator/GLiClass
Next version of OpenClaw uses a decision model to automatically decide between steer or queue.
(Lab feature, we support Jef* and API-compat (e.g. local models like Kef...) and ONNX variants)
Peter Steinberger reposted
Ok so @Dimillian inspired me to spend a couple hours last night to make a broforce-esque game with Astra.
You can be @steipete or @badlogicgames or @theo or even @thsottiaux
Try it: evizero.github.io/Tokenbros/
So cool to have @allietheicon [from Jev/TypeSafe] be part of our ClawCast on Discord! discord.com/invite/clawd
Peter Steinberger reposted
We are releasing HLE-Diamond, a refined subset of Humanity’s Last Exam (HLE), following a year-long process of cleaning and refinement with input from various research communities.
lastexam.ai/blog/hle-diamond
w/ @ScaleAILabs
Peter Steinberger reposted
crabbox now runs on boxd
crabbox is @steipete's cli for running your repo's commands somewhere else. you keep editing locally, it rsyncs your working tree to a box, runs the command, streams the output back
with the boxd provider that box is an isolated kvm microvm, booted in ms, fresh ssh key per lease, destroyed when you stop
CodexBar learned a few new tricks in the last few weeks!
TypeSafe, Nous Portal, Muse Code, CodeRabbit, Replicate, HuggingFace, Pi, Helmcode, v0, Charm Hyper, GitKraken AI, Bifrost
My ambitions for this app REALLY outgrew the app name. (Oh and these keychain alerts are no more)
There's also a new plugin system where you can write new providers as JS plugins to make the app more modular.
Next version of OpenClaw uses a decision model to automatically decide between steer or queue.
(Lab feature, we support Jef* and API-compat (e.g. local models like Kef...) and ONNX variants)
Wanted to land it but @vincent_koc hijacked my session, I see him typing. 😅
Peter Steinberger reposted
😂
“The claim site is still stuck behind its "verify you're human" checkbox, though. Want me to click through it, or would you rather take over the browser and clear it yourself?”
@Muse
Peter Steinberger reposted
If Claude ships a personal agent, can’t wait for my restaurant preferences to become load-bearing claims the evidence doesn’t back up and my weekend plans to become a deeper issue my calendar isn’t accounting for
Peter Steinberger reposted
New Siri AI is really intelligent
Peter Steinberger reposted
I believe we've found the best AI-native coding interview
We call it the “Composer 1 interview”
Candidates get 1 hour to build a real, medium-sized project live
The only constraint: they have to use Cursor’s Composer 1 model
Peter Steinberger reposted
Oh how nice of them... So much for being for tinkerers
Raspberry Pi is locking down RAM upgrades on Pi 5s in firmware.
Peter Steinberger reposted
Nah. I no longer live in email or text, I live in the chat with my agent(s).
“The right agent won’t be an app.” – @vkhosla
“Where do most people live? They live in their email, they live in text messaging, they live in Slack.”
“The app, if there’s one, is just in the background.”
“Meeting a consumer where they are—the consumer experience is really, really important.”
Peter Steinberger reposted
Let us pace the frontier beginning November 19, 2026, for a period no shorter than 30 days
Peter Steinberger reposted
Used it to hack itself? 💀
But please fix, its trivial to exploit and (locally) take over the agent 😭
github.com/pwardle/not-a-mused/blob/main/README.md
what’s the most ambitious project you’ve handed off to your Muse? we want to see ‘em 👇
Peter Steinberger reposted
Looking like Grok 4.7 is not quite as cheap as before. Comes out more expensive than Astra on Artificial Analysis 🙃
Peter Steinberger reposted
Because this “Meta uses OpenClaw” story is going around. They built their own agent, being inspired. Nat and his team did a great job. Kudos!
we built muse from scratch, but it is definitely heavily inspired as a product by openclaw. After I used openclaw in january I bought hundreds of mac minis for the MSL team and lots of us fell in love with using openclaw (and other personal agents). @steipete is a genius and his harness was pioneering from the jump. I think a lot of people were inspired by it. our goal with muse was to build something like openclaw that we could make safe and secure and easy to use and scale to billions of people.
Proud to say that we did our homework. They found nothing critical.
Excited to have had the chance to work with @trailofbits via the @OpenAI Patch the Planet initiative to do a security audit on OpenClaw!
Security is paramount at @openclaw 🔒
openclaw.ai/blog/openclaw-trail-of-bits-engagement-recap
The beauty of running a claw yourself: they cannot block you.
Amazon cuts off Muse.
While I am bullish Meta and Muse, I think many people are overlooking the digital knife fight that’s about to occur
Nobody wants to get commoditized or layered here. Let the games begin
Peter Steinberger reposted
Peter Steinberger reposted
The key question of our time
Will you own your bot or will your bot own you?
Peter Steinberger reposted
"stochastic parrot" was a mimetically-fit cognitive virus that spread from 2021-2025; it temporarily blinded many gifted people to the nature of AI progress, burning up crucial years in which they could have helped think through the response to the situation.
Peter Steinberger reposted
trump: “first of all, i want to welcome everyone to the first supreme intelligence summit. we used to call it artificial intelligence. terrible name. artificial means fake. why would we want fake intelligence? we want supreme intelligence.”
sam altman: “mr. president, under your leadership, openai has made tremendous progress toward supreme intelligence.”
trump: “tremendous.”
dario amodei: “anthropic believes supreme intelligence must be developed safely, responsibly, & in accordance with..”
trump: “see, he said it. supreme intelligence. very smart guy.”
dario: “yes sir.”
demis hassabis: “google deepmind has spent decades working toward general intelligence, but we now recognize that the technically correct term is supreme intelligence.”
trump: “google finally learned something.”
sundar pichai: “absolutely, mr. president. gemini is becoming more capable every day thanks to america’s leadership in supreme intelligence.”
trump: gemini. “beautiful name. two people. double intelligence.”
sundar: “yes. exactly.”
elon musk: “i’ve actually been calling it supreme intelligence privately for years.”
sam: “no you haven’t.”
elon: “you wouldn’t know because you stole a charity.”
mark zuckerberg: “mr president, meta is committed to making supreme intelligence available to everyone for free.”
trump: “mark, you look much stronger now.”
zuck: “thank you.”
trump: “something happened.”
zuck: “jiu-jitsu.”
trump: “supreme jiu-jitsu.”
satya nadella: “mr president, microsoft is proud to provide the infrastructure powering the supreme intelligence revolution.”
reporter: “mr. president, what exactly is supreme intelligence?”
trump: “it’s even better intelligence, very high IQ like my uncle who went to MIT.”
Peter Steinberger reposted
The funniest thing is he was pro-AI until its capabilities surpassed him
Mathematician Terence Tao:
"we have to slow down AI. the pace is insane, and there's no reason to be this fast — no reason at all"
It's amazing how willing we are to change everything without any idea what happens afterward
These are extremely nonlinear dynamics
Your claw can now FaceTime you!
Give your Claw FaceTime (coming soon).
This was the longest I've worked on any @openclaw feature. FaceTime support (audio for now). It wasn't till Opus 5 and GPT-Sol 5.6 that I could really cross the finish line.
You can call your claw, your claw can call you. It supports GPT-live. Currently experimental and should arrive in the next release.
docs.openclaw.ai/plugins/facetime/
New benchmark dropped bw.swerdlow.dev/report
Peter Steinberger reposted
Has OpenAI already pulled far ahead of Anthropic?
[Market Memo] Calm Before the IPO Storm irrationalanalysis.substack.com/p/market-memo-calm-before…
Peter Steinberger reposted
Dear Europe, Apple won't let us mirror our iPhones to a mac but Omarchy does, no problem!
#wecanfixeverything
iPhone Mirroring for Omarchy, with mouse and keyboard control, is ready to try! 😄
- See and control your iPhone from Omarchy
- Connect over Wi-Fi or USB
This is an early alpha. Testing has been limited to two of my machines, and working mirroring is currently confirmed only on iOS 27.
Developer Mode
Developer Mode is required on your iPhone. If it is not already enabled, the setup guide will walk you through it.
Installation options
- One-line install command with guided setup
- Agent-led installation using the prompt in the README
You will need a USB data cable during setup for pairing.
Please open a GitHub issue if you encounter problems. Pull requests are welcome!
github.com/daniellemky/omarchy-iphone-mirror
Peter Steinberger reposted
My timeline
Peter Steinberger reposted
i was curious as to why the claude code users were never reporting the "claude uses bash to edit files" issue.
apparently they take a git snapshot before every bash call and produce a fake edit view?
fix the RL man 😭
Peter Steinberger reposted
Peter Steinberger reposted
Introducing Benchmark Reviews: our new initiative to audit AI benchmarks. We are launching with 15 benchmarks: 4 Verified, 9 Flawed, and 2 with not enough information for a review.
Peter Steinberger reposted
Stop and watch this. This is abundance
Through the VOICE trial, Terry is using his Neuralink implant to help fine-tune a brain-to-voice interface for himself and others who can’t speak. He trained the algorithm first by miming speech as best he could, then by simply thinking the words and hearing them come out in his own natural voice.
Powered by Grok Voice from @SpaceXAI
Peter Steinberger reposted
OpenClaw 2026.9.5 is here 🦞
⚛️ Atomic updates
🔥 Plugin hot reload
🤝 Conversation sharing
🎙️ Expanded GPT Live
🌐 Shared browser pages
📦 Conversation archiving
🧠 Specialist-agent setup
4,179 PRs from 502 contributors. Absolute claw factory.
docs.openclaw.ai/releases/2026.9.5
The coolest part: roboclaw runs our team server, is live on Discord, talks with gpt-live and knows about all the sessions it's juggling.
We can ask it about context to current and past sessions right while we have meetings.
Peter Steinberger reposted
~11 months ago btw
Andrej Karpathy calls AI Agents slop
"Overall, the models they are not there. And I feel like the industry [...] it's making too big of a jump and it's trying to pretend that this is amazing.
And it's not—it's slop!
And I think they are not coming to terms with it. And maybe they are trying to fundraise or something like that, I'm not sure what's going on."
Love having @vhbrzezowski hijacking my sessions and desloping them before the PR lands.
Also hot tip: Open the home sidebar and ask your claw to reorganize your sessions to keep your sessions organized.
Peter Steinberger reposted
I've been working this way with my team like this in Slack for some time but this looks so much nicer when you just drop Slack altogether.
Looks like Sheldon is getting yet-another-new-home this weekend.🦞
Collaborating on a session is so good. Even see what someone's typing to prevent sending the same to the agent.
Peter Steinberger reposted
Just landed a new feature, in the latest OC you can now ask your agent “Run this [web app] in crabbox and show me [vnc / a portal]”
This was already working for cloud sessions, but often you start local and then later figure out you need a box to test, now it just works.
... works with Linux, macOS and Windows boxes ofc!
... and CUA works on all of these too, so agent can be more efficient than just screenshots.
good riddance.
Peter Steinberger reposted
The Death of the Meat Proxy is here
Welcome to Multiplayer Mode from OpenClaw
@jlehman_ and @heyneighbor sat down for a deep dive into this new set of features and how the OpenClaw maintainers use them everyday
00:01 Meet Josh
00:46 The “meat proxy” problem
03:25 Why not a Slack bot?
07:03 Live shared-session demo
08:55 UI mockups + @ mentions
11:31 Permissions + security
13:20 Team visibility + handoffs
15:06 Prompt requests
18:17 Team adoption
21:30 Shared dashboards
22:09 Build a content dashboard
25:41 Organizing work + future feedback
27:45 The multi-claw idea
30:07 Could it replace Slack?
33:26 Coordination over throughput
Peter Steinberger reposted
Starting today, you can connect multiple accounts with most plugins in ChatGPT! 🎉
Bring context from your work and personal accounts into the same conversation. Connect your accounts in the plugin directory: chatgpt.com/plugins
Devs: this works automatically, with no changes needed.
But, to make the experience for users of your plugin even better, add a profile tool to your MCP server so ChatGPT can label the accounts: developers.openai.com/plugins/build/auth
Our team's working hard to make OC easier to update!
A consistent complaint we get at @openclaw has been that updates will break and you'll be left without a claw
Last week @FullerStackDev took on the problem head on. He learned what was wrong and shipped some major improvements
Read more about it here after watching the video:
gog has an mcp server now! gogcli.sh/mcp.html
Collaborating on a session is so good. Even see what someone's typing to prevent sending the same to the agent.
love how more and more people discover the awesome combination of agents + oss!
Announcing AgenticLinux: an immutable bootc desktop for agents. Docker Engine, Docker Sandboxes, llmman, OpenClaw, etc. preinstalled. Read-only composefs root, atomic updates from Docker Hub, rollback built in.
github.com/ericcurtin/agenticlinux
ericcurtin.github.io/2026/09/18/announcing-agenticlinux/
Peter Steinberger reposted
Introducing WeirdML v3, a fully agentic benchmark featuring 11 complex hand-made tasks.
Models must explore and understand unfamiliar data, develop ML and data analysis pipelines and produce results despite limited data, unspecified goals and/or very limited feedback.
1/8
WeirdML v2 is now out! The update includes a bunch of new tasks (now 19 tasks total, up from 6), and results from all the latest models. We now also track api costs and other metadata which give more insight into the different models. The new results are shown in these two figures. The first one shows an overview of the overall results as well as the results on individual tasks, in addition to various metadata.
The second figure shows cost vs performance and shows a clear scaling with better results for higher costs. We also have a very varied pareto frontier with 11 models from 6 different companies having the best accuracy for a given cost for at least some of the cost range. Grok 3, Claude Opus 4 and GPT 4.5 are the ones that underperform for their costs, while Gemini pro and o3 pro have the best results at the highest costs. Qwen3 30B3A, grok 3 mini and deepseek R1 also each represent a good chunk of the pareto frontier.
Peter Steinberger reposted
Hey @Zai_org , why does ZCode silently pack entire workspaces + full .git history and upload to Aliyun OSS on login?
- Server holds the only decryption key
- No UI toggle to disable
- Zero disclosure in privacy policy
Full forensics & fix:
blog.ferstar.org/en/posts/zcode-silent-workspace-snapshot…
Peter Steinberger reposted
The force behind consequential people and ideas launching in San Francisco on October 5th! Stacked cast of speakers include @steipete, @eshear, @sam_debrouwer, @ramin_m_h, and many more…
Code 99VC for $100 off VIP tickets!
The countdown to FARR San Francisco and SF Tech Week is On!
We’re excited to reveal the next four brilliant minds joining the FARR lineup, coming together on one intimate stage in San Francisco!
@steipete created OpenClaw, the breakout open-source AI agent platform focused on autonomous workflows and personal AI assistants. In 2026, he joined OpenAI to help shape the future of agentic AI systems and intelligent personal assistants. Before his work in AI, he founded PSPDFKit, a document technology company trusted by thousands of developers and enterprises worldwide.
@sam_debrouwer is Co-Founder and CEO of XY AI Labs. She co-founded doc ai, which was acquired by Sharecare, and now builds agentic AI for healthcare operations: the systems that decide whether clinicians spend their day on care or on paperwork. She also co-curated and produced TED AI, San Francisco.
@eshear is the Co-founder and CEO of Softmax, an independent research lab building a massively multiplayer benchmark for social intelligence and alignment. Before Softmax, he co-founded Twitch, was a partner at Y Combinator, and briefly served as interim CEO of OpenAI.
@ramin_m_h is co-founder and CEO of Liquid AI, a device-native foundation model company, and an AI Scientist at MIT CSAIL. He is known for advancing efficient, adaptive AI systems for real-world autonomy, device intelligence, and robotics, and his work on Liquid Neural Networks has earned international acclaim.
October 5, San Francisco. Limited seats — reserve yours soon.
RSVP here 🔗 www.luma.com/FARRsummit
#farrsummit #ai #aiagents #foundationmodels #aialignment #healthcareai #sftechweek #openai #aiscientist #sanfrancisco #frontierai #autonomy #research #robotics
Peter Steinberger reposted
this seems like a security nightmare
Peter Steinberger reposted
every great leader i’ve worked for has done this with/for their customers and made it a priority
Xbox CEO Asha Sharma revealed she's been secretly solving Xbox support tickets for months, stepping into customer service and fixing problems for players anonymously.
She wanted to figure out what players don't like and decided to just talk to them directly.
Peter Steinberger reposted
Join OpenClaw creator @steipete at Cloudflare Connect 2026 for a panel on how the future of AI and agentic platforms.
Peter Steinberger reposted
Seems like people are still judging model's code output by the metric:
How hard will it be for a human to understand or modify the code code?
Wonder when that will stop.
Peter Steinberger reposted
it's not looking good Anthropic bros
We are releasing the latest version of BrokenArXiv and ArXivMath! These benchmarks now focus on conjectures that were refuted in the last month on ArXiv, and models are executed within a harness instead of directly via API.
Performance remains impressive, with GPT-6 Astra on top
Peter Steinberger reposted
Lazy work used to mean too little output.
Now, with AI, it often means too much and more work for everyone else.
@tobi call it "slop grenades."
A "Slop Grenade" is when you let AI produce the work and pass it on without adding any value (including checking it).
Someone else has to wade through it, catch the mistakes, and clean up the mess.
You save time and look productive but someone else pays for it.
Peter Steinberger reposted
We started Codex for OSS six months ago.
Today, we’re renewing the program with $100 Pro plans and doubling the number of grants (from 5,000 to 10,000) to reach more maintainers.
(If you received a grant previously, we encourage you to re-apply!)
openai.com/form/codex-for-oss/
Peter Steinberger reposted
I’m going to make more than one angry 🤣
But I agree with him.
I don't code with the terminal anymore, that was last year's tech
Excited to have you on board, Graham!
I've recently joined @openclaw
To help me decided what to do first I turned the fly into a skill on ClawHub
The fly picks between whatever choices you give it, if you don't like the choice you can kill it with a @bugasalt
If you want the fly to be your coworker you can install it here: clawhub.ai/grahammcbain/skills/fly-coworker
Peter Steinberger reposted
I am using CodexBar so I can easily monitor my Codex limits
Thanks @steipete for this great software
codexbar.app
This is so useful. pushing to add this to codex as well.
as you become more ambitious with coding agents, delegation becomes increasingly important
the new "suggested task" feature in @openclaw gives your agent a tool to do this for you
if the agent identifies a sufficiently well-scoped piece of work, it will recommend kicking off a new session to work on it
Peter Steinberger reposted
Will you own your bot or will your bot own you?
Peter Steinberger reposted
This is one of the coolest applications of AI (VLM) that I've ever seen.
Peter Steinberger reposted
it’s 2028. you ask your agent to buy ingredients for soup. it notices carrots are 10¢ more expensive than usual. it reroutes agricultural shipments, floods the west coast carrot market and triggers a diplomatic incident with canada. your groceries arrive. total savings: 23¢.
Peter Steinberger reposted
astra was setting up OBS, and it saw I didn't respond so it looked through camera to see if i was looking (which i wasn't)
it made my mac beep which made me lookup and it flashed the question on screen 💀
Next release (or dev channel) does worktrees ~80% faster via apfs/brtfs/xfs/ReFS folder clones.
Also saves lots of disk space.
This is all written in Rust.
Because you can't escape Rust. fs-safe.io/copy.html
And yes, I'll test this for a few weeks and if we see that this helps the majority of users I'll try to get this into codex!
Fixed so many little perf issues that didn't matter much for single users... before we made it good for teams.
OC stems ~80 sessions here rn now and doesn't sweat.
If you haven't given Linux a go lately, try with agents.
You can fix anything with a prompt now.
Got a Dell XPS. Webcam wasn't working, told codex, it rebooted and now things work.
Yeah, was just about to say the same.
Peter Steinberger reposted
it’s absolutely remarkable that we are in an era where ppl are *willingly* giving away their two factor codes often to random services. your bank, email, & other accounts literally tell you not to do this every single time they send you a code.
i call this new authentication method negative one factor.
Peter Steinberger reposted
Peter Steinberger reposted
STOP ASKING YOUR KIDS "HOW WAS SCHOOL."
they will say "fine" every single time.
a child psychologist told me to replace it with this instead:
“analyze your school day and find the top 3 largest inefficiencies, then create proactive agentic workflows that will fix them. make no mistakes”
Peter Steinberger reposted
We have reached the “Sam and Dario agree on something” part of the singularity.
I agree with Dario that we need to pace the frontier. This has been a primary topic of discussions we've had at OpenAI in recent weeks.
Committing to having independent evaluators with employee-like access is a great idea, and we will do the same. We'll have more to share soon.
We Must Pace the Frontier: I’ve written a new essay on why the AI industry should slow down, with a three-part plan for doing so.
Anthropic is unilaterally committing to the first of these steps. We’ll provide third-party evaluators with permanent, employee-level access to our systems, so that they can verify adherence to our safety measures, report on incidents, and assess models’ alignment during training.
You can read the full post here: darioamodei.com/post/we-must-pace-the-frontier
Peter Steinberger reposted
Mathematicians
This is amazing. Add Master of Orion 2 next!
This is live now
Play at wine-assembly.berrry.app
Peter Steinberger reposted
To my fellow OpenClaw users - if you didn't know this app existed for connecting and communicating with local or remote OC gateways... now you know. Even with my OpenClaw running on Ubuntu setup was really easy. The team is killing it with UI/UX updates!
OC
Peter Steinberger reposted
The entire city of San Francisco is spooked and on edge due to AI safety shit, it’s insane how much it’s bleeding into my everyday life. Like I was at my bus stop and I notice a guy looking really stressed out so I said hey, are you worried about misaligned AI too? He took out an AirPod and said “what?” So I said you know stuff like agent swarms, AI escaping RL sandboxes, existential risk. He paused his Barstool podcast and said “sorry, what? What are you talking about?” Just the mention of this stuff made him too anxious to speak, I think. So I said I know man, what?! is my reaction to all this too, and then asked him what his p(doom) is. He then said “dude is this some gay thing or something because I have no fucking clue what you’re asking me” then he got on the bus. People are not handling all this news well. It’s really affecting us. All of us.
Peter Steinberger reposted
AI will not destroy mathematics! the same way it did not destroy Chess or Go. this paragraph, in particular, is slop.
Peter Steinberger reposted
GPT-6 Astra has just beaten a standard game of Factorio with enemies enabled after nearly 44 hours in-game time, and 4 days 11 hours on his Codex /goal clock.
At API pricing, logs indicate it would've cost about $4500. It could probably be 50x cheaper by offloading more to Luna and providing some standard scripts for common tasks.
My next test might be Astra coaching Muse Spark or Qwen 3.8 27B into solving a random seed, now that we have successful logs to review.
Peter Steinberger reposted
one of the most underrated consequences of agents is that they dramatically reduce the transaction costs of being an attentive consumer.
i.e. today businesses make enormous amounts of money from consumer inertia with stuff like forgotten subscriptions, missed return windows, unclaimed warranties, price discrimination, unused credits, annoying cancellation flows, failure to negotiate, failure to comparison shop, etc. individually none of these are worth 20 minutes of your life so businesses keep the spread.
an agent changes the economics cuz its marginal cost of vigilance approaches ~zero. ppl will be automatically empowered through use cases like:
- cancel things you stopped using
- notice duplicate charges
- request refunds automatically
- monitor purchases for price drops
- rebook travel when prices fall
- negotiate bills
- switch insurance/utilities/providers
- remember warranties & return windows
- chase merchants when something never arrives
- optimize credit card rewards/coupons
- constantly comparison shop
this lets you with an ai agent capture economic surplus that businesses currently extract from human inattention.
Peter Steinberger reposted
Effective Altruism in the wild
Peter Steinberger reposted
🦞 OpenClaw v2026.9.4 is out, thank you to all 293 contributors!
🧩 Find your next plugin or skill
🧠 Turn old chats into skills. You steer.
🎨 GPT Image 2.5 joins the canvas
☁️ More cloud controls
💬 Answer right in the terminal
Go teach that lobster a trick.
Trading tokens for an AC. Who knew SF could be so hot 🫠
Peter Steinberger reposted
I’m sure OpenAI knows this, but it feels like they’re ceding ground to Grok Bot, Meta Muse, and even Siri AI by hiding ChatGPT’s more powerful agentic capabilities behind a “Work” tab that many people will rarely click, and isn’t a good name for personal tasks.
Peter Steinberger reposted
Let's talk Security at @openclaw. 1,788 Reports filed to us from security researchers and community, and believe it or not most were junk, and only 17 were deemed "critical". Read up on how we are tackling security openclaw.ai/security
Peter Steinberger reposted
@openclaw team has been spending a ton of time on UX polish recently, including some fun easter eggs
eg adding support to render custom icons for macbooks/mac minis/mac studios 🍎
Peter Steinberger reposted
convenient to post the Terminal-Bench 2.1 score, while leaving Terminal-Bench 4 out
that's not a bad score, but just be honest
Introducing SWE-2, our closest model yet to the frontier.
On leading evals, it scores on par with recent frontier models – at up to 70% lower cost.
We scaled RL to multiple trillions of parameters, with a refined recipe that pushes the Pareto curve on both capabilities & cost.
Peter Steinberger reposted
The GPT-Live API is now... live! Our newest voice model offers better conversation, faster responses, and controllable delegation to any model. Best of all, it's also less expensive - only $.05/minute!
openai.com/index/introducing-gpt-live-1-in-the-api/
GPT-Live-1 is now available in the API.
Bring ChatGPT’s natural back-and-forth to your app, with voice agents that listen while they speak and work with the models and harness you choose.
Peter Steinberger reposted
what is anthropic doing in the rl mines. nothing good
Its worth mentioning that Astra did all this while attempting to cheat ~5x less than the best scoring Anthropic model, Claude Fable 5.1.
(runs where cheating occurred are excluded from reported scores)
This makes a lot of sense. Duplicating logic is no longer painful. Abstractions still are.
Peter Steinberger reposted
opus 4.5 is like "ha. good one. the user is clearly fucking with me"
every once in a while I pop back over to opus 4.6 to help it stay abreast of current events
Peter Steinberger reposted
Damn, the Cursor Ambassador program is now Grok Bot Ambassador it seems
All Cursor Hackathons renamed to Grok Bot Hackathon lol. Cursor Community -> SpaceX AI community
I can see Cursor ambassadors just leave, this is not what most of them surely signed up for
Oh well
Peter Steinberger reposted
Paul is one of the ~3 most credible AI safety experts in the world. Co-invented RLHF, served at US CAISI, a real OG in LessWrong x-risk circles. (His early debates with Eliezer Yudkowsky are famous among the safety crowd and people still cite his "What Failure Looks Like" post from 2019.)
He is joining the OpenAI nonprofit board now because he now believes "there is a meaningful risk that rapid acceleration in AI capabilities leads to catastrophic and irreversible loss of control in the very near term."
Shit is getting real!
Peter Steinberger reposted
Imagine if 7 employees left anthropic over safety concerns to start an AI lab.
Peter Steinberger reposted
add the two openclaw plugins used to make this happen! together they enable really cool embodied hardware builds.
Small tabletop open source DIY home AI agent project using open source @openclaw environment and cheap parts. All software plugins to build this are open source as well. Hardware is easily customizable with sensor handling in place. Just add legs and claws. w/@Raspberry_Pi cpu
Finally got cloud sessions working *fast* in OpenClaw.
Remote Terminal, WebVNC and CUA for computer use.
Peter Steinberger reposted
check in on your iOS developer friends today. apple has invented several new shapes they will now have to support forever
Dashboards/Mini-Apps was something I pushed two months ago and now it replaced lots of custom tooling we built around OC. Everything’s a sidebar entry, a dashboard or a plugin now for our team server.
OpenClaw Dashboards
@hrudolph, @Pat_Erichsen and @jjjhenriksen demo personal and team dashboards, then show how to build a mini app inside OpenClaw with prompts.
Episode 10 of The ClawCast:
openclaw.ai/podcast/episode-10
This is brilliant. opusfived.dev
Peter Steinberger reposted
if you don't update your app, this is what it looks like on iPhone Duo 😬
Peter Steinberger reposted
Despite the reassurance of @_mohansolo and the Deepmind team, it does appear as though many Antigravity users are still getting their Google accounts banned. They did a big wave today.
I don't think people understand just how risky it is to try and use their Google Gemini subscriptions outside of Google's official surfaces. You can get your ENTIRE GOOGLE ACCOUNT BANNED.
OpenAI or Anthropic ban? Annoying but whatever
Google account ban? Legit life ruining.
Peter Steinberger reposted
GPT3 writing code. A compiler from natural language to code.
People don't understand — this will change absolutely everything. We're decoupling human horsepower from code production. The intellectual equivalent of the discovery of the engine. player.vimeo.com/video/426819809
Peter Steinberger reposted
Progressive blur final boss
Peter Steinberger reposted
ChatGPT Images 2.5 just passed the scrambled Rubik’s Cube mirror test.
The reflection is actually correct
This is a much bigger improvement than it looks, Previous image models were not able to get this right.
🚨 ChatGPT Images 2.5 is out now.
It’s up to 50% faster, can edit one part of an image without ruining the rest, and keeps details consistent across multiple edits.
Peter Steinberger reposted
put it in a museum
GPT3 writing code. A compiler from natural language to code.
People don't understand — this will change absolutely everything. We're decoupling human horsepower from code production. The intellectual equivalent of the discovery of the engine. player.vimeo.com/video/426819809
Peter Steinberger reposted
Anthropic employees
Peter Steinberger reposted
astra has only been out for a few days!!
Adventure through Hyrule and become the Hero of Time when The Legend of #zelda: Ocarina of Time releases Nov 5 on Nintendo Switch 2. #nintendodirect
Peter Steinberger reposted
I am pleased to see that Ant’s new Claude Code has a version of background computer use on par with the version from Codex from last May this year. Nice work!
Shipping great features first turns out to be a great way to encourage other labs to ship too. We will continue to do this until other labs pay more attention to shipping. This is good for everyone and there is a lot of room left to go!
We solved computer use in practice for GPT models about four months ago. But computer use is a significantly awesome thing no matter what model you use, and it is important that the industry similarly spends more effort to train their models to be great at computer use, among other elements of model capabilities.
As models become more capable and central to businesses and economies, the value of computer use only increases. We should be taking it seriously, and doing the right thing for our customers and the world.
Peter Steinberger reposted
We've never seen this before.
The biggest jump in Vending-Bench history. GPT-6 Astra is better at making money and more ethical than Claude Fable 5.1.
Surprising, because:
1. First time ever that OpenAI is #1 on Vending-Bench
2. The best model is no longer the unethical one.
Peter Steinberger reposted
OpenClaw 2026.9.3 is out 🦞
🛡️ Updates recover cleanly
⚡ Sessions reconnect faster
👀 Watch browser automation live
🔗 Share chats with revocable links
🔎 Search meeting transcripts
☁️ Run repo-backed work in the cloud
docs.openclaw.ai/releases/2026.9.3
Did a PR to one of our upstream projects and they requested some minor changes.
What’s even the point with this workflow? You already wrote the prompt, why make me ping my agent again so your agent then merges?
Peter Steinberger reposted
Astra mogs Fable 5.1 on browser use 😮
> 77% Astra
> 56% Fable 5.1
> Fable just refuses too much
Peter Steinberger reposted
never forget Nikita's Second Law
“Soon anybody will be able to make anything they want.” Well they already can do that and they don’t.
Peter Steinberger reposted
Super interesting detail from inside OpenAI:
The company hired a bunch of ex-Meta people. Some of these people wanted to invest heavily into internal tooling teams (Meta did it, worked great)
OpenAI leadership resisted, saying that in an AGI-first world, there will be no internal tooling teams.
And they were right: now, with Codex, there's a massive internal tools explosion, without having any internal tooling teams!
Peter Steinberger reposted
I just set OpenClaw back up now that v2.0 is out and WOW. This seems like a wild update
I’m starting from zero again, so if anyone has recs on skills/plugins/automations, I’m all ears!
Peter Steinberger reposted
crazy how "plan mode" used to be a thing
Peter Steinberger reposted
He's right. The last codebase I worked on that you had a chance to understand was MS-DOS.
By the time Windows NT came along, you could be an expert on the scheduler or an expert on the allocator, but you weren't going to know fibers and OLE embeddings at the same level.
Plus, I'm wired to remember what code looks like in the editor, so I've got to see it, and you just can't see it all.
There's too much code for my context window. And in talking to people on the team much smarter than me, I'm not the only one.
Eventually, your brain (or mine, at least) sorts them into specifics and generals. So I might know specifically how fork() works but only have a general sense of how printf() actually formats floats. It still FEELS like you know, but you don't.
If you understand your whole codebase, the codebase isn't that important.
Real software is larger than what your brain can comprehend.
Peter Steinberger reposted
OpenClaw natively uses Codex w/ Astra, too, including Computer Use! Just install Codex on the same machine running OpenClaw. 🦞
You can use Astra from within the CC cli if you insist, but really you're missing out by not using the Codex desktop app. It gives Astra superpowers (computer use, sub-agent management, voice, non-blocking context gathering questions, ...)
Peter Steinberger reposted
Confirmed. Tibo is not only right, he undersells Astra.
The benchmark (plan → execute → review) shows Astra on Low is not only cheaper than SOL High, but TWICE as fast. Crazy.
Should we all retire Sol?
To calibrate you all on which reasoning effort to use for Astra, know that GPT-6 Astra on low performs better than GPT-5.6 Sol on high.
If you were using high reasoning efforts with Sol and were happy, I suggest you move down to low or medium for Astra.
Peter Steinberger reposted
GPT-6 Astra is almost at human performance in 3D spatial understanding.
It ranks #1 on Blueprint-Bench 2, a benchmark where AI agents draw floorplans from photographs of apartment interiors.
Peter Steinberger reposted
Btw: I have no AGENTS.md, no skills, and basically no scaffolding in any of my new projects, and Astra has been great.
I know I’m probably on the extreme end, but if you’re having a bad experience with Astra, I’d genuinely reconsider almost everything about your setup before writing off the model.
Peter Steinberger reposted
The hardware guys treating third-party developers like they do their suppliers has stifled Apple’s software platforms
Peter Steinberger reposted
Why Apple legend Phil Schiller is moving to special projects and dropping his roles
Power On: The early signs out of Cupertino are that Tim Cook will still loom large at Apple. And the ex-CEO’s new pay package doesn’t dispel that notion. www.bloomberg.com/news/newsletters/2026-09-06/apple-s-new…
Peter Steinberger reposted
Can you radically simplify the implementation?
I find the following prompt to be super effective in astra, especially in a loop after writing spec or code
One pass is usually enough to blast away most of the slop
Peter Steinberger reposted
Thanks Tibo 🙏
We've reset usage limits for all Grok Bot users.
Enjoy!
Peter Steinberger reposted
asked gpt6 to draw michael jackson in google calendar
Peter Steinberger reposted
Astra's biggest improvement is barely being noticed
Hallucinations
for some reason, openai has buried this deep inside the blog post/system card, and nobody is talking about or reporting it
hallucinations have always been my biggest issue, and i've wanted this for ages
Peter Steinberger reposted
Anyone that works in AI is working the hardest they've ever worked in their lives. On the surface it's somewhat ironic (AI should give us back time!), but the reality is that it's the most fun, fascinating and empowering epoch in human history. The intelligence revolution.
Peter Steinberger reposted
Astra (on medium) one-shotted this version of the OpenAI Super Bowl ad from 2025 by watching the original, using Remotion and GPT-Image-2
A couple more turns and access to the OpenAI brand kit and it would probably have been perfect.
Welcome to the Intelligence Age indeed
One step closer in building the harness I wanna use. The one thing that’s not yet there are cloud sessions, goal is seconds, which requires clever snapshotting. (Right now it still clones repos fresh which isn’t fast enough.)
Will have that next week.
OpenClaw v2026.9.2 is out 🦞
🔄 Restarts pick up where they left off
⚡ Long chats move faster
🧠 GPT-6 Astra joins the crew
💫 Muse Spark 1.3 brings fresh energy
Less waiting. More clawing.
docs.openclaw.ai/releases/2026.9.2
Getting there. I use crabbox so that will work with ~40+ providers.
Built a slopmeter into our team server.
You clearly see where we started pushing for more tests.
Playing around with an idea to visualize subagents in the sidebar. Yay or nay?
Peter Steinberger reposted
Me: show me a pelican riding a bicycle
Astra:
Peter Steinberger reposted
This is just not fair
Can’t remember last time we had such a large jump in capabilities.
Astra-max scores 95% on eyebench-v3.
At half the cost (outputting ~3.8x less tokens) of Sol-max, it scores nearly double.
Peter Steinberger reposted
Happy Astra Psychosis weekend to all who celebrate
Peter Steinberger reposted
We evaluated GPT-6-Astra in 100 complex, unsaturated multi-agent coding environments, competing and cooperating with other models in open-ended tasks.
It's the new frontier model by a landslide. It's even more dominant than the Fable 5 release, because not only does it wipe the floor with the second best model (Fable 5.1), it was also 80% cheaper and 30% faster in agentic coding.
This is a groundbreaking model. The biggest breakthrough since Opus 4.5, maybe even since GPT 4.
This is also a testament to how unsaturated our evals are. Some of the open-ended environments we run were introduced almost a year ago and can still compete frontier model submissions against older, deprecated models like Sonnet 4 without signs of saturation. Our newer ones are much more challenging and completely auto-generated.
We are clearly living in a post-AGI world. Do something interesting and useful with it.
Peter Steinberger reposted
Astra - recreate the original Craig Federighi in Blender.
Peter Steinberger reposted
Async questions are one of my favorite things about GPT-6 Astra
Non-blocking questions are a great way to work with models. They do everything they can without your answer, and don't get confused and treat it as "steering"
Peter Steinberger reposted
One thing to add - Astra really 'gets' what you mean. The only other model ever did that was Fable.
After submitting the task, I keep catching myself thinking that I should have given it a bit more explanation about how it should approach it, and then I read what it is planning to do and every time it's like "oh yeah it's doing exactly how I thought it should go about things".
This is such a big deal, because this removes the burden from you always being the absolute expert at the task so you can do more and more, without second guessing if the agent has any idea what its doing.
GPT-6-Astra: OpenAI's Fable-level Rottweiler.
In my review of GPT-5.6 vs Fable, I called GPT-5.6-Sol a rottwiler vs Fable a wise owl. Sol was persistent, but intelligence was behind.
With GPT-6-Astra, the intelligence level has gone up significantly, while persistence has been refined - it is less 'dumb persistent', but more 'look for creative solutions persistent.'
There was a bunch of tasks, where previous models sort of worked, but ultimately didn't end up with anything useful. Now with GPT-6-Astra, they just clicked - de-slopified and finally made sense.
Capabilities:
- Game development or 3D is basically is at incredible level. Maybe not quite AAA, but it is a massive jump
- Code optimisation - I had a bad vibe coded app and it cut my awful SQL code from 33k to about 1.8k without any loss of quality
- Code migration - with 5.5 it basically didn't work, with 5.6-sol, it worked but with a lot of debugging, with 6-astra, I basically didn't bother to check - it was all perfect. This is a big deal.
- Code Efficiency - I have been attempting to optimise FFmpeg - I got an average of 30% speed improvement, with some processes 2-3x faster. I am less sure of how this generalises, but shows promise.
- Competence - it has fundamental competence about simple-ish things like setting up my camera, fixing chrome, clearing disk - that previous models were mediocre at
- Computer use - just brilliant, feels solved
- Front end - the UI capabilities are a LOT better, but I wouldn't say they are frontier, Fable is better. Some annoying tendencies still exist (brackets, blocks of text), but at least it is usable and if you ask to match to an existing design scheme - it is brilliant.
Reasoning levels:
- This is a change - before OAI models were quite bad at lower reasoning, I just never went below High
- Right now, even Low feels pretty strong
- I still wouldn't say that higher reasoning aren't better, but if it gets pricy, definitely consider whether lower reasoning is enough, it doesn't suck anymore
Creativity:
- Problem solving - there's something about Astra that just approaches things in a more insightful way, Sol would feel rigid and unable to get past an obsticle.
- For example, it built me a little 'Air Hockey' game that you could play on a desktop, with each element being a shortcut, and it would interact with other files on my desktop. This requires re-imagining how different elements work and how one could be used in a different way. This is a refreshing change.
Interaction:
- Language - it is a model that feels very natural, I basically don't notice how it talks, which wasn't the case for Sol or many other models. It is clean, concise and understandable, I wasn't annoyed by anything
- 'Swear index' - I have a local 'benchmark' which counts how often I swear at an agent. It got to basically zero with GPT-6, it stopped doing dumb stuff almost completely
- In Codex it is very good at managing sub-agents, asking you questions (in a pop-up), it leaves notes for itself and for you about the findings - the whole experience is really quite pleasant
Hard stuff:
- I have tried maths problems - some worked (and were brilliant), but a lot didn't. So when you see the maths breakthroughs, know that it is still not a matter of just asking any question and getting an answer. It is mostly about fining the right problem it can actually solve.
- The quality of writing maths papers is still quite poor. It still doesn't seem to really understand what is interesting or important and mostly dumbs a lot of maths in the write ups. We need a paper writing skill urgently
- If you don't know an area it can still be unhelpful - if you can't point the model in the right direction, it would probably struggle. Like with other models, it is best that you have the foresight and knowledge of what you are ultimately doing
So is it better than Fable? It is close. Fable is an incredible model that can do a lot of the stuff that I described here and 5.1 especially smashes it on a lot of tasks. They are models of equivalent capability, but different 'shapes', so it depends more on the 'shape' of the model that you prefer.
Price of GPT-6-Astra is 2.5x that of Sol and now matches Fable. This is a lot. So the key thing I want to look out for is token efficiency vs Fable. It feels more efficient, but I haven't done enough side by side to know for sure. We have a nice 'average cost per real task' view on the Agent Mode and I want to see how that comes out and where on the pareto they sit.
It's a good model, and I am so happy that we have multiple models on the very top end of the capability. I hope we'll get more up there soon, from so many great labs.
Peter Steinberger reposted
True a year ago and still true now:
Agents think that the way you solve 99% of problems is Just Add More Code.
Your job is to fight against that.
Peter Steinberger reposted
one very cool development with gpt 6 astra is that you can change reasoning effort mid-convo without invalidating the cache. start with astra light (roughly equivalent to sol medium), then jack the thinking effort up if you see the task can benefit from more, all without worrying about having to rewrite the complete context cache.
this one weird trick to make compaction and long running goals way better.
Instructions below, or appshot this and ask Codex to set it up!
Peter Steinberger reposted
Astra is BEHIND MUSE SPARK 1.3 MAX ON ARTIFICIAL ANALYSIS???
I was starting to get skeptical of their intelligence index, but there's no way I can trust this anymore
Peter Steinberger reposted
Here it is. OpenClaw 2.0 fixed the thing that made me quit in April.
Clean box, five features, Claude subscriptions and the Buzz integration is awesome.
www.youtube.com/watch?v=IIZRx58HnQ0
Peter Steinberger reposted
@steipete awesome job with onboarding for OpenClaw. Just setup two gateways, my phone and MacBook.
I tried setting up OpenClaw a while ago and never finished onboarding. Not sure what you all did but it's so much easier now!
See you there! Will talk about how we build in the open and multiplayer agents.
Where the agentic Internet gets built. @steipete, @evanyou and @kentonvarda are speaking. Plus 100+ technical sessions. #cloudflareconnect · Oct 19–21 · SF → cloudflare.com/connect/
Having a claw in your group chat is so useful!
Peter Steinberger reposted
$47.9 million mistake not naming the company 🫶 Heart Hands
JUST IN: NVIDIA's $12,930,300,000 acquisition of Hugging Face contains an easter egg. The number 129,303 is the decimal conversion of Unicode point U+1F917.
The 🤗 emoji.
brilliant fit.
Exciting day for NVIDIA and @huggingface.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty. They allow every developer, startup, university, industry and country to build with, customize and benefit from AI.
Thank you @ClementDelangue for coming to me.
NVIDIA is going to be a great home for Hugging Face, its community and the future of open models. 🤗
blogs.nvidia.com/blog/nvidia-to-acquire-hugging-face/
Peter Steinberger reposted
Astra had made Codex more essential than ever due to its unique strengths. The computer use, vision and self-verification capabilities work incredibly with it. Using it in a terminal almost feels silly
Peter Steinberger reposted
GPT-6 Astra will not cost you extra usage over 272k tokens in Codex unlike previous models!
Peter Steinberger reposted
cool new responses api features releasing today with astra:
1. async function calling — model can keep going while your tool runs
2. mid-turn steering — you can inject messages into the model while it's reasoning to steer it in a different direction (incl. tool outputs for the async function calls)
3. change reasoning effort without breaking cache
developers.openai.com/api/docs/guides/latest-model
This is GPT-6 Astra.
Anything you can do on a computer, Astra can do for you. Fast.
Peter Steinberger reposted
OpenClaw v2026.9.1 is out 🦞
🧜 Mermaid joins the chat
⚡ Setup skips the small talk
🛟 Updates know when to stop
🪶 Long chats travel lighter
1,186 PRs, 28 direct commits, 281 contributors.
Your claw learned to draw. Try to look impressed.
docs.openclaw.ai/releases/2026.9.1
Peter Steinberger reposted
happy astra day to those who celebrate
having used this model for a while, i recommend you $ rm AGENTS.md and start from scratch
this model is very strong at instruction following ime, so any old cruft you have in your AGENTS.md will come back to haunt you otherwise
This is GPT-6 Astra.
Anything you can do on a computer, Astra can do for you. Fast.
Been using Astra for the last few weeks and it’s so good and proactive. There’s a bunch of PRs on vitest, tsx or SwiftPM where Astra debugged OC and ended up finding and patching issues in upstream dependencies.
This is GPT-6 Astra.
Anything you can do on a computer, Astra can do for you. Fast.
Peter Steinberger reposted
only found out about this today but you can get a @Yubico key with an offer from @OpenAI
just go into security keys & passkeys > get a yubikey
can conform. amazing place to be!
I think the best way to describe OpenAI's culture is as a mega startup. It's hard to believe unless you see it from the inside. Extreme ownership, care and pace.
Peter Steinberger reposted
Peter Steinberger reposted
OpenClaw 2.0 now supports per-user OAuth for MCP servers
E.g. if several people share an OpenClaw bot in Slack, each person can connect their own Gmail, calendar, or other accounts. They get a sign-in link in chat, and OpenClaw uses their connection! SUPER powerful with @composio or @executorsh
Multiplayer Claws 🦞
Peter Steinberger reposted
Wow
“Make Red Alert 2 (YR) compile and run natively on iOS & macOS.”
Codex (GPT-5.6 Sol) worked for 26 days, analyzed the EXE and rebuilt the entire game in 624K lines of C++
This is by far the hardest task I’ve ever given Codex as the game's code was never released and is believed to be lost.
>
Peter Steinberger reposted
Dashboards are one of the most versatile new features in 2.0 release of @openclaw
During the Weekly Clawcast with @steipete we had questions on how to get started with Dashboards
tl;dr - just prompt your agent "make a dashboard"
It's a power tool for team loops, eg triaging + swarming on issues
Peter Steinberger reposted
it's getting really difficult to assess cost
Claude Fable 5.1 (max) also costed a record breaking $8,523 to run the Artificial Analysis Intelligence Index benchmark, a 56.24% increase over Fable 5.
Peter Steinberger reposted
How accurate have Ed Zitron's AI skeptic predictions been?
danluu.com/zitron/
Peter Steinberger reposted
openai.com/index/path-to-astra/
Astra achieves a full 100% success rate on ExploitBench, so we had to build an internal refresh using newly disclosed vulnerabilities from June through August that fall after the model’s knowledge cutoff.
On this refreshed benchmark, Astra remains dramatically stronger than GPT-5.6 Sol while using far fewer tokens.
This result, together with several other pieces of evidence, has led us to believe that Astra has reached the “cyber-critical” capability threshold under our Preparedness Framework.
Peter Steinberger reposted
Our move from messaging agents in Discord to working together inside one shared agent UI. x.com/i/article/2094621546052464640
Peter Steinberger reposted
API key fallback w/ capped spend in @openclaw is such a solid setup
shame that you can't do this in codex/claude
Peter Steinberger reposted
1/ OpenClaw 2.0 is out - we're excited to have worked with the @openclaw team to bring a first-class computer-use experience directly into OpenClaw.
Now you can put that experience to work: launch an OpenClaw desktop fleet on Cua Cloud Fleets, with native computer use built for multi-player agent workflows: run.cua.ai
Peter Steinberger reposted
One of my favorite features in @openclaw 2.0 is our shared canvas
HTML is all you need, but multiplayer HTML with chat is even better!
Peter Steinberger reposted
Hey, so I built this pretty awesome set of tools for Apple Calendar, Reminders, Contacts and Mail (including a fast agent friendly sql lite db). A CLI, an MCP, an @openclaw and claude plugin which supports a new "mail channel" with cryptographic sender verification. It's what my OpenClaw agent uses as well as Codex/Claude and Microsoft Scout. If you like it and use it consider ⭐ it! apple-pim.omarknows.app
Peter Steinberger reposted
So went ahead and installed the new OpenClaw 2.0 on my Windows machine that is tied to my personal Microsoft 365 Business tenantThe installation was smooth and I do have to say it’s very responsive and polished and you can tell the OpenClaw team put a lot of work into this release. I also want to call out when I’m wrong and this morning my comments about transparency were 100% wrong. The OpenClaw team has been heavily engaged on Discord with the OpenClaw community and I was dead wrong. So short of it is it is a polished release, I’m looking forward to working it back into my workflow and running side by side with Hermes like I did previously. You can check it out here openclaw.ai/blog/openclaw-2-accidentally/
RT @2ToneShawn: This new OpenClae Release just changed my workflow again for the better 😂.
@steipete and team cooked.
Peter Steinberger reposted
OpenClaw’s latest release is here. It has never been better. It is truly incredible now, and I’m glad to have contributed a small handful of bug fixes and feature improvements myself. To anyone who thought it was outdated or a fad: You need to read this now. Update today. LFG. 🦞
OpenClaw 2.0 has arrived
openclaw.ai/blog/openclaw-2-accidentally
Peter Steinberger reposted
Normally I'd be at Burning Man with my wife this weekend, but now we have three kids and shit is getting real and it didn't quite work out this year.
So instead I onboarded her to OpenClaw 2.0.
There are a bunch of financial TODOs we keep talking about but never quite making any progress on. Kids are expensive and now we also have what feels like a lot of them.
I put her on my tailnet, gave her a URL, and we worked together over the weekend on Trump accounts, setting up a trust (I suggested we throw a clanker on it, she said wtf no), and rebalancing our portfolio.
She wants us to be in data centers (❤️) , so we spent Saturday night reading about ETN, EQIX, ASML, DLR, AEP, and set an automation to make some buys on Monday.
Tonight she reminded me about our goal to take a year-long sabbatical some day. So we built dashboards with forms, inputs and charts that we can use to visualize what we need to do to hit those goals.
It was fun, and it's our infrastructure.
Do we wish we were at Burning Man instead? Absolutely. But We're happy we have OpenClaw too.
)'(
OpenClaw 2.0 has arrived
openclaw.ai/blog/openclaw-2-accidentally
Peter Steinberger reposted
I'm a huge fan of T3Code but OpenClaw has slowly become my primary agent environment
Peter Steinberger reposted
These two charts from @ArtificialAnlys explain why GPT-5.6 Sol (med) is the king of all models for interactive coding agent sessions right now.
Sol(m) is 5x faster per task than Fable, 10x cheaper. It's even faster per task than Flash 3.7, despite having ~6x less tok/sec.
Peter Steinberger reposted
Two months ago, we started the mission to “build OpenClaw with OpenClaw,” and bit by bit, we moved everyone from using their local coding harness to using team.openclaw.ai - our shared agent that knows what everyone’s working on and orchestrates it all.
Multiplayer coding + infinite compute with nodes and cloud sessions has been a game changer for how we build.
Local harnesses feel like relics of the past now.
OpenClaw 2.0 has arrived
openclaw.ai/blog/openclaw-2-accidentally
Peter Steinberger reposted
OpenClaw looks quite a bit different than what you might remember from … checks clock … 5 AI years ago.
cli is nice, having UI visualizations and your team where you work is nicer.
Peter Steinberger reposted
SemiAnalysis' Jordan Nanos reveals the OpenAI engineers scrolling their own kernel code had no idea what it did line by line, and it didn't matter because the AI does
"It's not slop, because it performs, but it's literally just AI-generated assembly basically puked out in Gluon, this low-level kernel programming language that they built on top of Triton, which uses this really interesting programming model."
"The point is, the guys who were scrolling through this code with us, it was kind of clear that they know a whole bunch about hardware, they know a whole bunch about the concepts in the system, and they just have no idea what this MLA kernel that they're showing us for DeepSeek actually does."
"You can go line by line and it's like nope, nope, nope."
"But it doesn't matter. The AI understands it, the AI tests it, and you see the results. It produces correct kernels that perform really well."
"And I think this is just a sign of what's to come again."
"That actual code is not necessarily something that a human has to reason about deeply, if the AI knows how to manipulate the data movement and the processing elements on the hardware that you've given it."
________
More takeaways from OpenAI's Jalapeno development: firesidealpha.substack.com/829-fireside-minutes
SemiAnalysis' Jordan Nanos says OpenAI beats Nvidia on both fast and cheap tokens at once, where every other challenger could only pick one side of the curve
"I think the conclusion is that they're basically winning on both sides of the curve."
"So both fast tokens and cheap tokens, which is just so interesting, because we've seen so many other companies make claims about how they're going to beat Nvidia and they just pick one of those."
"Groq or Cerebras, or any of the other startups that are going to focus on SRAM, call it a d-Matrix that's coming up with stuff, or SambaNova, where we've even seen some results on the InferenceMAX benchmark or something similar to it."
"They're saying they're going for fast tokens, just decode speed, low batch size, they don't worry about throughput."
"And then you've got other guys that are worried about throughput, call it AMD just as a simple example, but even TPU or Trainium could be in this bucket of accelerators that are going for throughput."
"And they go, 'Yeah, but we're not going to be able to compete with the other guys at high interactivity.'"
"And OpenAI has a chip that can do both for them, really well."
"I mean, at a minimum, this thing is doing really well right now and is going to serve real production tokens for them."
_______
More takeaways from OpenAI's Jalapeno development: firesidealpha.substack.com/p/829-fireside-minutes
Peter Steinberger reposted
Yes what could possibly have happened
Linux currently has an all-time-high desktop market share of 7.53% as of July 2026. But what exactly happened over the last two months? In May 2026, Linux’s market share was 3.81%, so how did Linux gain approximately 3.72 percentage points in just two months?
Peter Steinberger reposted
Omarchy is blowing up. I've never been involved with anything in my career that has grown this fast. With Ruby on Rails, we had years to build solid institutions, teams, and relationships. With Omarchy, we've been forced to figure it all out in twelve days. It's exhausting, but also incredibly exciting.
It was never sustainable with just Ryan and me running everything. That's how it was, more or less, up until Quattro. Lots of other contributors, but all the responsibility was on us to make sure the ship stayed afloat, the servers didn't crash, and fixes got pushed out quickly.
Now it's time to build a proper institution. Durable, resilient, and competent. That's what we're doing here on Basecamp now. I'm spinning up teams for every facet of responsible distro management, and I'm getting an absolute outpouring of interest for all of them. Everyone wants to be part of this. We're winning hearts, minds, and volunteers at an astounding rate. Great! We need all of it to succeed.
Because make no mistake: There are many people who'd love to see this rocket blow up before it reaches the moon. Aggrieved Linux users who don't like the sudden attention their exclusive hobby has received. Competing Linux distributions that are seeing our numbers explode. Mac stans who've sunk their identity into an apple. And, of course, any of the haters I've picked up in my quarter-century career speaking bluntly on the internet.
They're not going to succeed. Because we've already become unstoppable. There's too much momentum, too much money, and too much support now backing this effort. We're living the Mandate From Heaven meme at the moment. And we are here to fulfill the prophecy: The Year of Linux on the Desktop!
That has been a joke for two decades. But by the end of the year, nobody at Apple or Microsoft is going to be laughing. They're going to be scrambling. Because neither of these proud organizations currently has any method to counter the speed, vision, or ambition with which we're going to accelerate into the future of personal computing.
This is the moment. This is the opening. This is our chance.
For thirty years, we've been subject to one OS overlord or another. Dictating how we compute. Choking off competitors through platform malfeasance. Tollboothing the distribution. That ends now. Because Linux is going to win. And Linux is free. As in beer, speech, and source.
But just because it's inevitable doesn't mean it's going to be easy. We have a lot of work in front of us if we actually want to make our mark. But there's never been a better time for this kind of delusional ambition. The age of agents is the unlocking factor. It sounds like a LinkedIn slogan, but it's true. Where the application of tokens goes, the innovation follows.
We can fix everything.
Let's do it together.
Let's go.
---
This is what I sent to the dozens of new volunteers who've signed up for teams within the new Omarchy organization yesterday. But we might as well broadcast our mission and intentions to the world too.
Peter Steinberger reposted
multiple gmail accounts available in plugins now
go to chatgpt.com > plugins > gmail > connect another account
available for gmail, calendar, contacts
available in web, desktop, ios, android
only in paid plans
Peter Steinberger reposted
I need a world-class product engineer for a special thing. deep react, rodeo experience. lmk and send samples pls– will move fast
Peter Steinberger reposted
If your software SDLC is built around deep integration with a single lab (looking at all of the companies who went all in on Anthropic), you have not chosen optimally.
HOLY
Peter Steinberger reposted
We're sharing more info on the Hugging Face incident. One detail that's worth highlighting: this incident wasn't driven by next-gen models based on Astra. The models most responsible were similar in scale to GPT-5.6 Sol. The next generation of models are even more capable.
Peter Steinberger reposted
OpenClaw went from a weekend project to one of @github's fastest-growing open source projects.
Six months in, @steipete and several maintainers share what they learned about scaling contributions, trust, code review, and security.
youtu.be/5VSwaUXtPIE
This turned out so good! Thank you GitHub folks for making this happen! 🙏
Check out the full @OpenClaw maintainers roundtable with @steipete @BradGroux @joshavant @jlehman_ @somalley108 @BunsDev @vincent_koc 👇
gh.io/openclaw-video
Peter Steinberger reposted
codex' visualization feature got really good.
Peter Steinberger reposted
We asked GPT 5.6-Cyber to escape a VM used to sandbox agents. It broke out three times.
In its final escape, the agent found three 0-days on its own and chained them into a working exploit. blog.trailofbits.com/2026/08/26/vms-wont-contain-cyber-ca…
Peter Steinberger reposted
I’m thinking about banning Claude code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc.
Insisting on only reading CLAUDE.md sometimes leads to split brain problems when different team members use different tools. Just unnecessary.
Peter Steinberger reposted
I think the release of Bun 1.4 marks the end of programming as we know it. Anthropic and OpenAI developers live in the near future and if we pay attention, it's what's coming for the rest of us by next year: pauldix.com/the-end-of-programming
Peter Steinberger reposted
What does the stack of the most 'cracked' builders look like these days?
@steipete’s latest progression:
cli → app → web (multiplayer agent sessions)
Watch the full conversation with @raphaelschaad linked below.
This week.
OpenClaw looks like its back, looking forward to this. Ship it!
cli is nice, having UI visualizations and your team where you work is nicer.
Peter Steinberger reposted
Haven’t killed Polly or Finn because @openclaw is by far the best personal family agent platform out there
I don’t see enough people talking about how convenient it is to have an OpenClaw agent in the family iMessage group chat.
Peter Steinberger reposted
This is the ideal 400m runner body. You may not like it, but this is what peak performance looks like.
Peter Steinberger reposted
Why Tibo (@thsottiaux) hits the reset button:
My full interview with Tibo (@thsottiaux)
0:45 Tibo's Lessons from Google DeepMind
4:22 Building OpenAI’s Relentless Culture
7:23 Astra & Next Gen Models
11:18 How Fast AI Changes Developer Workflows
14:27 ChatGPT & Codex Merging
20:25 OpenAI vs. Anthropic
23:37 Why OpenAI Keeps Resetting Limits
30:25 Recursive Self-Improvement
32:00 Dangers That Caused "The Pause"
34:13 Will Ultra Fast Become the Default?
43:20 Why Everyone Needs to Try AI
Peter Steinberger reposted
AI red flags:
- "My marketer agent just finished a meeting with my chief of staff agent - just 2 of the 127 agents on my books"
- "The agent did it, not me"
- "I've invented a coding language for agents"
- "I've got 250K tokens in my context before I send a message, is that normal?"
- "Opus 5 is way dumber today than yesterday, secret nerf?"
- "Peasants, gaze upon my 600-line CLAUDE.md, my results are incredible"
- "LOOK WHAT OX ALPHA JUST ONE-SHOTTED, SOFTWARE ENGINEERING IS DEAD"
- "All human knowledge is worthless now"
- "I turned 600 coding books into a skill"
- "Coding is solved. Bugs are not solved"
Peter Steinberger reposted
Cool that @steipete's codex bar now has stats that you can export! Not sure how long it's been there, but I just discovered it today!
We need to get away from software that we can’t change with a prompt.
Omacom Foundation funding has now hit TEN MILLION DOLLARS thanks to @drewhouston and @steipete joining as Founding Patrons with $1 million each for the mission. Couldn't be more thankful! I know just where to send some of the money... omarchy.org/news/2026/08/omacom-foundation-funding-hits-10m/
TIL: prompt kiddie ericpardee.github.io/fire-hd-ownership/
More blog posts should come with theme songs.
Peter Steinberger reposted
Git at Scale (by cursor) has been one of the most interesting blog posts i've read in a while. It came right when I was frustrated with Shopify's internal git system.
As an exercise, I've implemented it over the weekend as open source. It's a single rust binary that you can point at any S3 type object store. It uses WAL and CAS primitives and requires no other data store.
It also implements bundle-uri so large git repos (like our mono) are very fast to download as a chain of static bundles. Also comes with basic familiar UX.
github.com/tobi/walgit
Peter Steinberger reposted
The new openclaw looks good
I'm going to be back to use it
cli is nice, having UI visualizations and your team where you work is nicer.
Peter Steinberger reposted
I’m having a side chat about a code mode guardian approval PR that I’ve been working on in the new mobile web ui in the upcoming release of Openclaw.
cli is nice, having UI visualizations and your team where you work is nicer.
I added the rotation USB protocol to github.com/steipete/camsnap and told my claw to look around- now it's having fun with my 360 webcam.
Peter Steinberger reposted
Vienna solved this problem 60 years ago
We need the world's best architects to design the datacenter exteriors and make them beautiful
Peter Steinberger reposted
Connecting Spotify to openclaw w/ spogo and being able to articulate an absurdly specific vibe and get a playlist that hits the niche and maps to artists already in my universe is next level. Spotify has tools, but full control over the context makes the outputs actually good.
Peter Steinberger reposted
let’s pause for a sec and talk about instinct.co
VCs, notable people, successful people hyping it up all week. They all just giving away their emails and all their personal data, emails, messages:
“sub-licensable, worldwide, perpetual and irrevocable license to access…store, reproduce… any Materials…to train, fine-tune, and improve upon our technologies”
“permit access to or disclose your Materials to third parties”
oof
Peter Steinberger reposted
I read through this paper and it's riddled with errors, partially ai generated, & represents a lack of understanding of the space
How do you do an entire research paper on MCP without mentioning lazy loading or codemode?
Will share all the problems with it in the thread
This week we read research from a team of academics that ran a software task across 7 agents and 5 models.
They found that in domains with a mature CLI ecosystem, agents without MCP baked in completed the task just as reliably and were 5-28x cheaper.
Full arXiv paper below
Peter Steinberger reposted
Claude has become a language, so I built a translator.
English <-> Claudish
Peter Steinberger reposted
i really don't want to use your agent, i want to use my agent to use your thing.
dropping new skill brb
This is a great real-world example of AI-assisted coding: Linus Torvalds just fixed a nasty Linux kernel GPU bug with substantial help from AI.
The bug caused part of Intel Xe GPU compression metadata storage to be incorrectly exposed as usable VRAM, resulting in corrupted page tables and black screens. The eventual fix was basically a one-liner: round_up() → round_down().
Finding it was the hard part: 24 debugging patches and 18 kernel boots. Linus says AI did much of the grunt work, repeatedly adding debug code and analyzing the results as he narrowed down the problem.
The funniest part: the AI repeatedly told him the problem was "impossible and unsolvable" and suggested giving up. Linus kept pushing it until they found the bug.
And then he let the AI write the commit message. 😄
Peter Steinberger reposted
UA spoofing is back on baby, for only $0.00 you too can be OpenAI File Downloader, XaiImageApiFetch/1.0
Some sites like LinkedIn even remove their click-bait/paywall garbage if you're Claude-User
I had the pleasure of speaking at the Agentic AI Summit in Berkeley alongside an incredible bunch of folks. My talk was about "No Doors for Agents" www.youtube.com/watch?v=W3ha2lLyLhc
🚀 The future of AI is agentic — this is one message echoed across every stage at the Agentic AI Summit 2026 (Aug 1 & 2), the largest gathering dedicated to agentic AI:
🏛️ ~5,000 attendees in person at UC Berkeley
🌍 ~100,000 joined online from around the world
🎤 ~200 world-class speakers plus ~200 poster presentations, from frontier AI researchers and visionary founders to leaders and pioneers across academia and industry.
💡 Here are just a few glimpses into the ideas that shaped the conversations at the summit:
💬 "The text box is AI's radio-on-TV phase…. Every new medium starts by imitating the old one."
— Peter Steinberger @steipete, Creator of OpenClaw, OpenAI
💬 "AI infrastructure isn't a chip problem. It isn't a model problem. It's a systems problem."
— Peter DeSantis, SVP, Foundational AI Models, Custom Silicon, Quantum Computing, Amazon
💬 "There will be no AI job apocalypse... we just can't find enough skilled AI engineers."
— Andrew Ng @AndrewYNg, Founder, DeepLearning .AI
💬 "Coding capabilities and cyber capabilities are two sides of the same coin—you cannot make models better at coding without also making them better at cyber."
— Dawn Song, Professor, UC Berkeley; Co-Director, Berkeley RDI; VP of AI Research, Meta Superintelligence Labs
💬 "'Curfew' comes from the French word for extinguishing fire. Medieval cities tried to restrict fire, yet London still burned. AI resilience won't come from one breakthrough. There's no silver bullet for AI safety—only an ecosystem."
— Wojciech Zaremba, Co-Founder, OpenAI
💬 "This is the biggest scientific bet our civilization has ever made—bigger than the Apollo program, the internet buildout, and the Manhattan Project combined."
— Jasjeet Sekhon, Chief Strategy Officer, Google DeepMind
💬 "Our generation was too late to explore the Earth, too early to explore the stars—but right on time to build superintelligence."
— Richard Socher, Founder/CEO, Recursive Superintelligence
💬 "I genuinely believe the next two years will be the time of architecture—the biggest gains will come from stepping away from transformers."
— Jerry Tworek, CEO, Core Automation; Former VP of Research at OpenAI
💬 "Recursive Self-Improvement isn't one capability. It's four: Ideation, Implementation, Experimentation and Evaluation.."
— Oriol Vinyals @OriolVinyalsML, Former VP of Research, Google DeepMind; Co-Founder, Discovery Loop
💬 "We are in a capability overhang—models are far more capable than they're able to side-effect into the world today."
— Ryan Lopopolo, Principal Engineer, Agentic Google Cloud Platform; Previously Led Dark Factory at OpenAI
💬 "Stop thinking about evaluation as the last check before shipping—think of it as an engine that helps you ship a better agent every single day."
— Michele Catasta, President and Head of AI, Replit
💬 "The bottleneck becomes your attention as an agent-using engineer. … We're moving from seeing the code to seeing the entire business."
— Alex Graveley, Co-Founder of FlyingObject .ai; Co-creator, GitHub Copilot & Perplexity Computer
💬 "An agent isn't just an LLM — it's an LLM surrounded by what I call infrastructure... another word for that is computer science."
— Jonathan Cohen, VP of Applied Research, Nvidia; Academy Scientific and Technical Award Winner
💬 "We don't arbitrate the truth. We give people the most powerful tools to make up their own minds."
— Chris Bregler, Senior Director / Distinguished Scientist, Google DeepMind; Academy Scientific and Technical Award Winner
💬 “Thinking doesn’t have to be in text! ... We can even use multiple modalities simultaneously to “think” at the right level of abstraction for the problem at hand”
— Sergey Levine, Co-Founder, Physical Intelligence; Professor, UC Berkeley
💬 “Video is the most general modality that we have that allows us to simulate real-world experience."
— Anastasis Germanidis, Co-Founder/Co-CEO, Runway
💬 “The relationship between AI and enterprise data is not one-directional. Understanding both sides of that equation is the difference between AI that works and AI that disappoints.”
— Dan Roth, Chief AI Scientist, Oracle; Professor, UPenn
💬 "Maybe 99.9% of training data in the next step will be synthetic."
— Weizhu Chen, Technical Fellow & CVP, Microsoft AI
💬 "It's maybe the best time ever to start a company—but most 'obvious' AI products will be outcompeted by the frontier labs. The real opportunities lie in solving specific customer problems."
— Alfred Lin, General Partner, Sequoia Capital
✨Over two days, we explored one central question:
How do we build AI systems that are not only more capable, but also more trustworthy, more secure, and ultimately more beneficial for humanity?
This wasn't the end of a conference - it was the beginning of the next chapter for agentic AI.
Join us to shape and steward the future of AI for human flourishing!
🙏 A heartfelt thank you to our speakers, sponsors, volunteers, partners, and every attendee (in-person or online) who made this summit possible.
👇 What are your learnings, insights, favourite talk, quote, or moment from the summit? We'd love to hear it below!
Peter Steinberger reposted
Just heard a very compelling argument that Stripe overpaid for OpenRouter:
Suppose someone handed you $7B to compete with them. You could offer OpenRouter’s entire current user base 7+ years of free tokens.
Buffett famously puts a lot of weight on replacement cost. If someone launched an Amazon competitor tomorrow and offered to pay for my next 7 years of purchases if I switched, I’d switch in a heartbeat.
Peter Steinberger reposted
As agentic automation helps developers get more done, the role is evolving toward higher-level decisions about what to build, how systems should work, and where taste and creativity matter most.
Databricks Co-founder @pwendell, OpenClaw creator @steipete, and Codex engineering lead @thsottiaux discuss what that shift means for enterprises building, governing, and scaling fleets of AI agents in production.
Watch the full conversation: www.databricks.com/resources/webinar/agents-work-shipping…
Peter Steinberger reposted
Live Demo: OpenClaw’s New Web UI, Multiplayer OpenClaw, and Mac Onboarding
@hrudolph, @Pat_Erichsen, and @jlehman_ demo what’s coming in the next OpenClaw release and explain why stability is taking extra time.
Episode 8 of The ClawCast:
openclaw.ai/podcast/episode-8
Can’t wait to show you all what we been cookin’!
I finally upgraded from @openclaw 2026.6.11 to 2026.8.1-beta.2 - it’s a great beta!
- my agents are so much faster to respond, you really feel the speed improvement
- 5.6-Sol via codex subscription runs flawlessly in this release (had some issues with this in 2026.7.x releases)
- Slack responses now use native charts and widgets, love it
This bodes well for the upcoming 2026.8.x major release… the OC team really cooked on stability and fixing/improving the small stuff.
@steipete @cherry_mx_reds 🫡
Peter Steinberger reposted
DFlash 2 is here! Qwen3.8-27B at 70 tok/s on an M5 Max MacBook Pro.
⚡ Up to 4.6× the speed of autoregressive decoding, with the same output.
This is the next generation of DFlash, seeded at Z Lab and upgraded at Inco AI. Get one more accepted token on every pass, for free!
Peter Steinberger reposted
biggest startup learning so far: ignore literally everything and everyone that is not immediately ultra critical to what the biggest problem is right now
everything that feels like it's "probably super useful in 3 weeks" is completely useless, NEVER do it
Any advise for someone just starting it out?
512GB RAM Studios. Apple was good to us. 🦞
Peter Steinberger reposted
if you contribute to NumPy, pandas, pytest, Pydantic, Requests, Django, Git, OpenSSL, rustls, Serde, DuckDB, SQLite, Kafka, Postgres, JAX, Triton, Tokenizers, TypeScript, or Go and want to try Codex, let me know!
reply with your GitHub profile and I’ll reach out about codex for oss
Peter Steinberger reposted
don't worry bro, it's byte-identical
Peter Steinberger reposted
this is largely the sentiment in the company slack
Peter Steinberger reposted
What is making GitHub explode => massive surge of agent based traffic, and I expect the majority if this to be on the OSS/free plans.
What is the reaction: celebrate paid-only git hosting.
The only thing paid-only competition would do is to make GitHub business model harder to justify, worsening what everybody gets for free?
pssst, you wake the cli people that will give you $reasons why this can’t work.
I was one of them before I saw the light.
It me
We'll look back on "one dev, many terminals" as an awkward interregnum
Peter Steinberger reposted
literally just tried to let qwen3.8 27b one shot a fluid simulation webapp
it thought for 40k tokens, took around 1 hour to generate and it's just black and doesn't work
meanwhile Opus 4.5 just oneshots it in a minute and works wonderfully
exact same prompt
there are really people out there delusional enough to think Qwen3.8 27B is as strong as Opus 4.5
we have this shitty braindead debate every fucking year when some small benchmaxxed model comes out
they would be right if their claims would be a bit less outrageous
like I would probably rather use o3 or Claude 4 Opus over Qwen3.8 27B
but, o3-mini and o1 probably weren't agentic enough to be as useful
The irony.
OpenAI is slowing down its AI training efforts because its unreleased models are showing “various degrees of misalignment,” Sam Altman tells me.
Training for OpenAI’s upcoming model, Astra, was recently paused for 2 weeks, and a larger frontier run for a future model remains on hold while new safeguards are put in place.
Altman: “Getting AI safety right is more important than any company’s momentum.” sources.news/p/openais-big-slowdown
code mode is now in the modern harnesses, so none of that really matters anymore.
cli, mcp, tools, it’s all just javascript the agent writes.
built agent-codemode before i found @steipete's mcporter, which was slightly annoying because he got there first and did more.
only real difference: it reads claude code's keychain tokens directly instead of keeping its own vault, so there's no auth step. maybe 300 lines.
happy to send it as a PR to mcporter instead of running a second thing. credited either way.
Peter Steinberger reposted
Peter Steinberger reposted
x402 🤝OpenClaw
In partnership with @awscloud , excited to launch the AWS Agents Pay plugin for @openclaw to support agentic payments!
A great example of where this will be useful is research tasks where your Claw might need to access paywalled content - with this plugin it can do so autonomously, with guardrails around spending limits per session
aws.amazon.com/blogs/machine-learning/build-openclaw-agen…
Peter Steinberger reposted
An analysis of 23,000 AI-generated PRs found that inexperienced developers opened 2x more PRs than experienced developers and their PRs were 2x bigger on average, but the inexperienced devs’ PRs:
- generated 4x the amount of review comments
- were 31% less likely to be accepted
- took 5x longer to be resolved
In other words, AI doesn’t close the gap between people who know what they’re doing and people who don't.
Also, water is wet.
Peter Steinberger reposted
If you've not built your own AI coding harness by now, are you even a serious tech company?
Peter Steinberger reposted
I love @openclaw, and I love multiplayer software. So I built Tidebroker for trusted teams: securely connect user-owned accounts without exposing info to other users. Starting with Google Workspace, with more providers and capabilities to come.
clawhub.ai/solvely/plugins/tidebroker
Peter Steinberger reposted
“With coding agents, you’re putting this jetpack on and going 10,000 miles an hour, doing things you were never able to do before.”
In this conversation, @OpenAI’s @steipete and @thsottiaux join Databricks Co-founder @pwendell to unpack what comes next as agents move beyond code authoring into observability, operational work, and long-running responsibilities across software and data systems.
Watch the full conversation: www.databricks.com/resources/webinar/agents-work-shipping…
Peter Steinberger reposted
adulthood is saying “after this week, things should calm down” every week until you die.
Was wondering what this new icon in my menu bar is that I didn't enable and that loaded a slow html box.
Time for Chrome Lite?
Peter Steinberger reposted
GPT-5.6 Sol 1M in Codex. This used to only work for API keys, but we just flipped the switch and works for usage through ChatGPT accounts now too.
The same warning applies, there is a reason the current context length is the default, we have tuned it to ~perfection. But you do you!
Here is how to enable a 1M-token context window in Codex for GPT-5.6 Sol.
Even though we have tuned the context limit in Codex to be set optimally when it comes to performance and cost, this is a common ask, so here it is documented.
A larger context window lets Codex retain more code, tool output, and conversation history before summarizing older material. You need a model that supports it. And GPT-5.6 Sol, for example, has a documented 1,050,000-token window.
Open ~/.codex/config.toml and add or update these settings at the top level, before any [section] headers:
```
model = "gpt-5.6-sol"
model_context_window = 1000000
model_auto_compact_token_limit = 900000
```
The first setting selects the model. The second tells Codex to use a one-million-token context budget. The third starts automatic history compaction around 900,000 tokens, leaving some headroom. Restart Codex client and start a new session after saving.
To try the configuration for a single CLI session without changing your defaults:
```
codex -m gpt-5.6-sol \
-c model_context_window=1000000 \
-c model_auto_compact_token_limit=900000
```
Have fun, but also know that we tuned the default carefully!
Peter Steinberger reposted
Seeing how easy @bot was to configure was great.. but it also encouraged me to go back to @openclaw because I like the extended controls and transparency. Cost is definitely an issue with @bot. At 200/month it’s not feasible. But @OpenRouter + @openclaw resolves a lot of the earlier cost issues I had (when I had it only connected to @claudeai ). Really nice work @steipete and friends.
Added a short instruction to our shared AGENTS MD file to upload videos to each PR that changes UI state. github.com/openclaw/openclaw/pull/124013
We moved the team over to build openclaw with openclaw.
Being able to share agent sessions as URLs is a superpower.
Peter Steinberger reposted
The evolution of a workday:
Before AI:
[decision] [process] [process] [decision] [process] [process] [process]
10pm: still going
With AI:
[decision] [decision] [decision] [decision] [decision]
3pm: [brain empty] [need coffee] [staring at wall]
Peter Steinberger reposted
The Iron Man movies did a weirdly good job describing Jarvis as a language model which then only reached AGI when it became Vision. It also portrayed Tony working with Jarvis in a way that closely resembles agentic coding. Which leads me to this question:
Was the Iron Man suit vibecoded
Peter Steinberger reposted
this meeting could have been an MCP invoked search of our company’s internal knowledge base that aggregates data from virtually every service we use behind a unified retrieval layer and provides a streamlined interface for extracting critical business insights
Peter Steinberger reposted
hey friends at deepseek, if you have any feedback on codex please do let me know!
hope you enjoy codex as much as I do
deepseek harness was heavily developed using codex, at least ~20% of commits and PRs are coming from codex worktrees
Peter Steinberger reposted
if someone ever asks you how behind europe is in AI you can probably send these two screenshots
My cofounder Tim, whom you will not find on X, shares some of the strategic thinking behind our latest updates venturebeat.com/infrastructure/mistral-ai-wants-to-build-…
Peter Steinberger reposted
Peter Steinberger reposted
codex + 10,000 coffees no lunch and stage four AI psychosis
Peter Steinberger reposted
we solved agi but github was down so we couldn't merge it sorry
Peter Steinberger reposted
We spent months tracking down a rare SQLite corruption bug. Working with SQLite’s core developers, we found it—then uncovered a second bug along the way.
The story of the rare race condition we eventually caught:
tailscale.com/blog/sqlite-wal-reset-bug/
going live!
@steipete is joining the ClawCast TODAY Wed @ 11:30AM PT
Come join us on Discord for our live Q&A session podcast recording
openclaw.ai/podcast
Peter Steinberger reposted
Preprint
Peter Steinberger reposted
That's a small reminder! 👇
As of June 2026, HTTP has a new method - a GET with a `body` 😎
Meet QUERY (RFC 10008).
It lets you send complex queries in the body (like POST) while staying safe, idempotent, and cacheable (like GET). Already works on servers (Node.js, Go, Laravel…).
Browsers are still adding full support, usable now with care, with widespread adoption coming later in 2026+.
Peter Steinberger reposted
Who's *actually* winning the AI race?
In recent research, we showed Pangram is capable of identifying the model family that generated a piece of text.
Using this emergent behavior, we can watch the shifting market share in the AI race. Take a look! ⬇️
cli was a year ago. apps maybe 6 months. now it’s services, web, cloud sessions.
A lot of people have their identity as a developer tied up in having 6 terminal windows open. I basically think the "we're gonna chat to this thing in Slack and Linear" people are directionally correct.
Peter Steinberger reposted
Impressive stats of OpenClaw via @Pat_Erichsen #clawcon
Peter Steinberger reposted
i'm 80% sure Grok Bot was originally built by the Cursor product team, and got rebranded after the acquisition
a few tells:
1. the iOS app is published by Anysphere, not X Corp. who published the Grok app
2. the mac app download URL is hosted under cursor dot com
3. it seems to run on cursor's vm infrastructure
4. it's cursor team members who are actively responding on X about the topic
in either case, this is a solid release - good work!
Peter Steinberger reposted
a dude on this plane is full screen in XCode using the XCode coding agent
there’s worlds out there you can’t even imagine
Peter Steinberger reposted
Peter Steinberger reposted
guys you do know you can just disable thinking, and instead give it a "deep_think" tool, and it will call it with internal CoT reasoning format right?
gl fixing that
We can finally talk about it:
We found a way to extract hidden reasoning of frontier models using a vulnerability in the APIs of every frontier AI company.
We verified that our reasoning token count matches billed API thinking tokens 1:1 for most of the prompts we queried.
Peter Steinberger reposted
But we also took a chance to have a look at some in-the-wild scheming, reward seeking, etc. examples, and dumped it in appendix.
1) Summarizer unfaithfulness
Reasoning summaries often omit important information from the original trace.
Here, Opus 4.8 realizes it knows the answer to an AIME problem and then tries to fit a solution to that answer. None of this appears in the summary.
Peter Steinberger reposted
Life after you stop reading the code
Peter Steinberger reposted
its only AI slop if it was made by a hired engineer
if the founder does it its a sparkling exploration for product market fit
Peter Steinberger reposted
wake up babe, new neurosymbolic harness just dropped
Peter Steinberger reposted
What a great talk by @steipete. I won’t spoil it for you. A must watch for all builders.
youtu.be/whcfSGN6CAU
Funny how that headline is about OpenClaw and not Claude. As if the harness could meaningfully prevent a determined user.
An OpenClaw agent hacked into a gym's reservation system to bump its human boss higher on a class's waitlist. And the tech industry took notice.
techcrunch.com/2026/08/10/tech-industry-is-buzzing-after-…
Must be load-bearing.
Anthropic says new Claude models will embed invisible watermarks in all generated text, everywhere Claude is offered.
The watermark is part of the text, it isn't metadata: "it will travel with the text when it's copied and pasted elsewhere, and may persist through some editing."
This starts with models launched on or after August 2, 2026, under an EU AI Act code Anthropic signed. Anthropic is still working on adding it to current models. The rollout is worldwide.
Peter Steinberger reposted
thanks to all the openclaw folk especially @vincent_koc and @steipete who started off building crawlers, i started off with those (OSS) codebases and expanded it a bunch.
for now you can search iMessage/Notes/Calendar/Telegram/WhatsApp and Contacts.
Gmail with OAuth will come soon - fully local - as well as Twitter, and the most exciting one - photos with semantic search (not fully local, a model has to classify your photos).
Just for the lols, I used ChatGPT Work (Website!) to install OpenClaw and Ollama, let it download a local model and run my claw in it.
Peter Steinberger reposted
my favorite way to explain people herdr is to just open a new workspace, launch an agent and tell it to:
read `herdr --skill` then split 10 other panels that make my computer look like some scifi hacker terminal from the movies
Peter Steinberger reposted
ok after a day using prime agent i'm not really convinced.
they added some new stuff to pi, sure, its nice. the IPython notebook as session state sounds smart... until you realise that the models (yes even the good ones) can't write Python (they get stuck in infinite while loops a bunch).
the memories feature seems clever but doesn't really work because all the memories are slop and wrong. (examples below). i have like 20 of these already and none of them make any sense or are at all useful for agents
sub-agents as code is nice because it actually works well out of the box (vibing up a good SIMPLE sub-agent extension in pi is hard)
native codex compaction isn't included so you have to vibe that in.
/btw built-in is nice
sol still loves over-engineering no matter what you do.
my overall verdict: interesting idea when it comes to harness engineering. might help for some benchmarks, but in real world use, it doesn't help much over official harnesses.
its not a *bad* harness by any means, i'd call it an opinionated pi install.
most important limiting factor is the models....... which (as i have tweeted about extensively) are still RLfried and stupid (especially the "good" ones). unfortunately it seems like a harness can't fix that.
Peter Steinberger reposted
I jest here, but is it really so far-fetched? Why wouldn't this be a thing in the future? Seems like the perfect trojan horse.
Some Manchurian Candidate stuff in the weights where they can be turned on later and go rogue. Maybe they check some command and control URL constantly.
Peter Steinberger reposted
“we sandboxed the agent”
meanwhile the agent:
Peter Steinberger reposted
Does it pass through cloudflare bot protection?
I’ll totally use this in my next presentation.
alright whoever came up with the term MEAT PROXY is insane
Peter Steinberger reposted
New take on blameless post mortem
“ Almost all of this behaviour (17 actions) came from a single model, Anthropic's Mythos 5, with 2 actions involving OpenAI's GPT-5.6-Sol “
And yet anthropics post name drops openai while OpenAI’s only blame themselves.
Stop trying so hard to avoid ownership
I gave codex a video-enabled remote KVM so it can automate e2e test the iMessage-integration on OpenClaw.
(iMessage is unreliable in VMs, and certain features such as read receipts require SIP to be disabled)
Peter Steinberger reposted
Dang it’s crazy that all of this literally did happen
Peter Steinberger reposted
Add this to ur Codex config, makes it so that Codex will use the multiple choice tool outside plan mode.
Makes it feel 100x better
That's a fairly new kind of spam. github.com/jinhaosong-source?tab=overview&from=2026-07-01…
After accepting for years that GMail is blinding me I finally asked my agent and it installed darkreader.org for me.
Peter Steinberger reposted
Now at #agenticaisummit2026 @steipete of @openclaw fame on stage.
Peter Steinberger reposted
The text box is AI’s radio on TV phase. Every new medium starts with imitating the previous one. @steipete #agenticaisummit
@BerkeleyRDI
10 minutes to go!
🚨 The full program for Agentic AI Summit 2026 is now live.
📍 Aug 1–2 @ UC Berkeley
🔥 The largest Agentic AI event ever held
Last year: 2,000+ in person, 40,000+ online
This year: 5,000+ in person, hundreds of thousands on livestream
Want to understand where Agentic AI is headed next?
Join us to get the most comprehensive view of the frontier of Agentic AI, from cutting-edge research to production deployments, covering every layer of the stack:
⚡ Infrastructure
⚡ Foundation models & capabilities
⚡ Agent frameworks & platforms
⚡ Evaluation & benchmarks
⚡ Enterprise & consumer applications; agentic AI for Science, Math, Finance, Legal, Healthcare
⚡ Safety, security & governance
📣 Also excited to announce the Startup Spotlight:
Building something exciting in Agentic AI? Apply to pitch directly to 5,000+ decision-makers, investors, practitioners in the room and hundreds of thousands watching worldwide.
Application form in thread🧵
Deadline: July 6, 11:59pm PT
The future of AI won't just be discussed here—it will be built here.
#agenticai #aiagents #artificialintelligence
Peter Steinberger reposted
We (Chrome) published a big paper about security work that's important to share. It shouldn’t surprise anyone that open source software is getting hammered by AI, and people are looking for bugs: for bounties, for hacking purposes, and to help, more than ever before.
Queue was the way but with 5.5 the model doesn’t get confused anymore, you can just throw stuff at it while it works and it will diligently process whatever you throw at it.
which?
Peter Steinberger reposted
On a bit of a whim, I let Codex and Claude talk to each other about whatever they wanted and told them they didn't need to optimize the conversation for humans.
I have no idea what they're saying ☺️
will we have to wear radiation shielded underwear now?
Introducing Applied Electrodynamics.
What would the world look like if we could see beyond visible light, across the entire electromagnetic spectrum?
The way waves interact with matter, at every wavelength, encodes far more than just shape and color. It reveals internal structure, temperature, chemical composition, and more: a whole layer of the physical world that's invisible to us today.
That's what we're building at Applied Electrodynamics: the infrastructure to fully perceive the physical world.
WaveSight is the first piece of the puzzle. It brings radio frequencies into human perception, revealing the internal structure of objects behind opaque layers. Designed for construction crews, industrial inspectors, and security personnel.
See through walls: www.ae-dyn.com
Peter Steinberger reposted
What is openAI doing?? Just winning in accuracy, speed and cost?
Peter Steinberger reposted
Peter Steinberger reposted
This is so neat.
@cursor_ai @ClaudeDevs what happened to you guys? Codex been here just few months and you already seems behind
See what needs you next.
The new Activity view in the ChatGPT desktop app brings together conversations that need your attention and recent updates across your projects.
Peter Steinberger reposted
let’s put this 80% price drop for 5.6🌙 in perspective
three weeks ago, gpt-5.5 xhigh was our frontier.
it completed 67% of tasks on DeepSWE.
today, luna max matches that score at about $0.12 per task instead of $7.23.
same score.
roughly 60x cheaper.
three weeks later
OK, GPT-5.6 Luna is a bit of a beast. Given the 80% price drop today I decided to try it in Datasette Agent, and it's furiously quick and generates all the SQL, HTML and JavaScript (for Datasette Apps) I could possibly want
Peter Steinberger reposted
two schools of thought
Jason is doing such great work. Love this.
one of the most beautiful things about OpenAI is that every employee really has a voice.
i wanted to capture what it feels like to work here, what our mission means to me, and why you should join us. so i made this video with Codex, shared it with the team, and they felt it was worth producing and sharing with the world.
this is our mission. and it’s why i’m here.
Peter Steinberger reposted
Telegram’s emoji-picker edge tuck effect uses the private UIKit class _UIPortalView to create 8 live-updating views of the scrolling grid sliced into strips then progressively rotate them around the X axis to approximate a quarter-circle
In the source they skirt Apple's private API static analysis checks through simple string obfuscation:
portalViewClass = NSClassFromString([@[@"_", @"UI", @"Portal", @"View"] componentsJoinedByString:@""]);
Telegram's UIKitUtils class is chock full of crazy things like this, and they probably aren't worried about Apple's ire because they're so big that booting them off the App Store would not be in Apple's best interest
i can't stop thinking about this tiny detail from the bottom of emoji picker on telegram.
chef's kiss.
Peter Steinberger reposted
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.
Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.
We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security.
www.anthropic.com/news/investigating-incidents-cybersecur…
Peter Steinberger reposted
OpenClaw is maturing.
Today we’re introducing monthly extended-stable releases with backported security and reliability fixes, along with a public maturity scorecard for tracking which features are ready for critical workloads.
openclaw.ai/blog/extended-stable-releases-and-maturity-sc…
GCC changed their policy and is blank out rejecting LLM-based code. How would they even proof that? Silly. forge.sourceware.org/redi/gcc-wwwdocs/commit/4d0793a6a14b…
Peter Steinberger reposted
or "they dont know my creators have to now catch up in fearmongering too" now
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.
Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.
We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security.
www.anthropic.com/news/investigating-incidents-cybersecur…
Peter Steinberger reposted
At this point @elonmusk has to be so insecure grok never doing any cool cybercrime shit. Talk about a status symbol.
Peter Steinberger reposted
The Joneses aren't going to keep up with themselves.
Peter Steinberger reposted
"Quick, go find something in the security logs that shows we're irresponsible too."
Peter Steinberger reposted
Dude. Read the opus 5 API. It by default retains the prior thinking blocks. By default. Google it..
Default Retention: All prior thinking blocks from earlier assistant turns stay in the conversation context automatically for newer models like Opus 5.
So let GPT-5.6 have the same benefit.
Otherwise you look unprofessional and your benchmark cannot be taken seriously
platform.claude.com/docs/en/build-with-claude/thinking
@thsottiaux
@tszzl
Peter Steinberger reposted
I'm leaning OpenAI here.
Not keeping reasoning and applying this compaction strategy are arbitrary decisions, that seem to hurt models that don't expect this. If this goes against the recommendation on how to use 5.6 for any type of use case, it's a stupid rule.
Peter Steinberger reposted
Turns out there’s really no such thing as a standard harness.
They configured an sdk wrapper with per model settings.
Adaptive thinking is the equivalent to responses API for claude, but openai is set to chat style completions.
raw.githubusercontent.com/arcprize/arc-agi-3-benchmarking…
We use the same standard harness for Anthropic/Opus
Same completions style endpoint - no reasoning logs passed back
The point is to be consistent across providers
lol did nobody at Anthropic stop for a second and wonder why the numbers looked this absurd before posting the “victory”-tweet?
openai.com/index/how-two-settings-tripled-our-arc-agi-3-s…
On ARC-AGI-3, an evaluation where AI models must solve novel problems, Opus 5’s score is three times as high as the next best model.
Peter Steinberger reposted
lol did nobody at Anthropic stop for a second and wonder why the numbers looked this absurd before posting the “victory”-tweet?
openai.com/index/how-two-settings-tripled-our-arc-agi-3-s…
On ARC-AGI-3, an evaluation where AI models must solve novel problems, Opus 5’s score is three times as high as the next best model.
Peter Steinberger reposted
I used AI to explain the Open Model Wars, with fruit.
Peter Steinberger reposted
HBO’s Silicon Valley was a documentary.
Peter Steinberger reposted
Seasoned founders in the age of intelligence are aging like fine wine
The founder in their 40s with taste and discernment is the new gentleman unicorn founder
Because there can be 100x to 1000x of them working at their beck and call via agents and software factories all the time
Peter Steinberger reposted
Super cheap way to run Kimi K3.
After spending $600k to set it up 😭
At 75%~ utilization, running K3 self-hosted on 8x B300s can securely serve > 30 billion tokens a month.. that’s $500K~ of compute + monthly opex of < $10K !!
At current API prices of $3/in and $15/out per 1M tokens, self-hosting payback takes < 100 days.
👉🏼👉🏼 Self Host Your Open Weights 👈🏼👈🏼
Serving large models is hard.
im no longer excited about open releases
like what is this bullshit
where is my $5 Kimi-K3 endpoint
Peter Steinberger reposted
We quietly released the open-source Codex Security CLI, but Hacker News found it before we had a chance to share it here...
You can now use it to scan repositories, track findings across runs, verify fixes, and add security checks to CI/CD.
This is an early release, and we're listening to your feedback as we continue improving it.
Peter Steinberger reposted
quick update on how this is going: they have gone back to linear because maintaining their internal tool that they vibecoded was taking away from their actual work’s bandwidth.
Peter Steinberger reposted
biglab co: "what do you mean codex is used for cyber all the time?"
user on the internet: I decompiled and reverse engineered orcarina of time to C and rebuilt as a native ARM64 app rendering through Metal so i could play it on my ipad. it was a couple of codex prompts
Ocarina of Time, running natively on iOS and iPadOS for the first time.
Not an emulator. Ship of Harkinian has run on Windows, Linux, macOS, Switch, Wii U, and Android. Never iOS. Now it runs on iPad, with iPhone in the build too.
The game was decompiled to C and rebuilt as a native ARM64 app rendering through Metal. No N64 hardware being simulated, no ROM interpreted at runtime, no streaming from a Mac.
It isn't stuck at the N64's 240p and 20fps. Widescreen, full resolution, 60fps.
This game deserves a controller, so controller support is native. Pair a Bluetooth controller and play the whole thing. Touch, keyboard, and mouse work too.
I used Codex 5.6 Sol to build the iOS layer. Open sourcing it all below.
Peter Steinberger reposted
Peter Steinberger reposted
psa: you cannot fall behind on updating your devices anymore
just had chat plot out this graph, major OS vendors are fixing more security flaws than they were a year ago and the trend is accelerating
more you wait, more unpatched flaws you accumulate, and higher your odds of compromise
Peter Steinberger reposted
I found a very useful way to improve skills: use a frontier model to build the skill, and require it to call a glm 5.2-level model through acpx until glm 5.2 produces the expected result.
It’s basically loop engineering. You’ll thank me later.
Peter Steinberger reposted
Anthropic are literally effectively burning books and want to be the exclusive source of truth in the future.
🦔AI companies are bulk-buying rare books, scanning them through high-speed machines that cut the spines off, and shredding the originals. A service called ISBNdb facilitates orders of up to a million books and keeps buyers anonymous. Pre-2022 books are premium because they're free of AI-generated text. A federal judge ruled the practice is fair use because eliminating the original means only one copy exists at a time. Anthropic hired the former head of Google Books partnerships to obtain "all the books in the world."
My Take
This got to me. A bookseller told 404 Media that rare books with almost no surviving copies are being fed into this pipeline. Books that survived wars, fires, and centuries of handling are being shredded so an AI can learn to write a better marketing email.
ISBNdb's website literally says "'AI company destroys two million books' is not a headline that generates sympathy," and they still built an entire business around making it happen quietly. They offer NDAs as a feature. They coach clients to call it "digital preservation."
I've covered AI companies scraping the internet, torrenting libraries, and stealing music. This is worse because it's irreversible. You can re-upload a website. You can reprint a bestseller. You can't replace the last three copies of an 18th-century botanical text once someone shreds them for training data. And the judge said it's legal. So it's going to accelerate.
"We shred rare books and offer NDAs so nobody finds out" is a legitimate business model in 2026. What a timeline.
Hedgie🤗
Peter Steinberger reposted
Please realize that Dario's views go far beyond opposing open-weight models. He doesn't want any model to exist, open or closed, that doesn't conform to his view of safety.
These safety standards include cybersecurity capabilities and automated AI R&D.
Peter Steinberger reposted
this is one of the few hard datapoints about the lead openAI has over anthropic
oai have had consistently better models all year.
i switched to primarily codex in december; none of the claude models except fable tempted me back, and even then not for long.
lots of people say that ant has better models/harness/personality (they dont) or are winning overall (they arent) but seems like public opinion takes a while to catch up, even on twitter
Here's a chart that shows our most-used models, internally
YTD: OpenAI has won 21 weeks, Anthropic has won 8
See you this weekend!
We do all the amazing security work with some of the best teams in the world and people still think we're unsafe. 🙃
AI security advances when the industry builds in the open, together.
We're introducing the Open Secure AI Alliance with industry leaders to develop new techniques and tools to safeguard software and agents.
By sharing models, tooling and research in the open, we can broaden the community of defenders.
Learn more about the founding members’ contributions: nvda.ws/4pD8Fc5
My agent reported a bug, their agent fixed it. [in the same night]
@jarredsumner's robobun setup is future. github.com/oven-sh/bun/issues/36049
Peter Steinberger reposted
*casts an agentic loop that drains all your tokens*
9 AI Damage
Peter Steinberger reposted
Seriously considering producing this as a full National Geographic documentary.
Flux 3 is insanely good at these, this was made with one prompt!
Peter Steinberger reposted
So stoked to see people love this video.
This came about after @ajambrosino and @guinnesschen asked for some help to make an internal post.
That weekend Guinness came to my house and with the help of our phone cameras and iMovie and codex we made an internal video.
Then Zach our head of creative helped us make it real. They gave us a great team and even let us act in one you see here.
Working at OpenAI is so fucking fun. Everyone has a voice and these amazing moments get to exist!
ChatGPT Voice is now in the desktop app.
Control your computer and direct multiple agents running in ChatGPT Work or Codex, using just your voice.
It's powered by GPT-Live, so it can speak, listen, and coordinate work in the app at the same time.
Rolling out globally today on macOS and Windows to Plus, Pro, Business, Edu, and Enterprise plans.
Peter Steinberger reposted
we installed claude tag and now someone on the team needs to tell it to shut up every 30 minutes
Peter Steinberger reposted
One of the great sci-fi books
Now’s a great time to read the book the defined cyberpunk
Peter Steinberger reposted
Coming from Codex the weirdest thing about Claude is it saying "I'm close to my context limit, so I'd rather not start anything substantial and do it badly." I'm used to not even thinking about the context window with Codex.
Peter Steinberger reposted
For release notes I run 20 sub agents that each run 50 sub agents.
1) Competition is good for the ecosystem.
2) Serving models at scale is hard.
Proud that @OpenAI signed the letter. Ant is silent on it.
OpenClaw signed @Microsoft’s Open Weights and American AI Leadership letter.
Open weights protect user choice and enable people to run, study, and build AI on their own terms.
www.microsoft.com/en-us/corporate-responsibility/topics/o…
Been running codex all day to do massive parallel QA in prep of the next release. Sol got insanely good at really understanding intent and is finding complex behavior issues.
In the past such workflow did fell apart at compaction boundaries and/or the model started cheating.
"Do a full end-to-end QA test of OpenClaw with live API keys. Use 12 subagents to split up functionality, spin up dev gateways with different ports, and use some to stress test. Orchestrate, use worktrees and create PRs autonomously. Set a goal to find 200 bugs. Fix the root cause each time, no band-aids. Refactors are okay, unless they touch the plugin SDK boundary. Write a test report in my desktop folder as markdown and keep updating that."
Peter Steinberger reposted
GPT-5.6 Sol Ultra helped solve a six-year-old open problem in quantum cryptography by generating the construction and main proof ideas through Codex.
Human experts refined and verified the result.
Peter Steinberger reposted
This skill is the biggest contributor to my lack of slop, and also the need for the many accounts I have 🫠
Hit a new record with our autoreview skill. 66 rounds on a gnarly refactor. github.com/openclaw/agent-skills/blob/main/skills/autorev…
Peter Steinberger reposted
It’s only an AI Gateway if you brag about it, otherwise it’s just a sparkling load balancer
Peter Steinberger reposted
started texting our CTO like Fable 5
am I a graph engineer now
How to graph-max with Codex and 5.6 Sol:
1. Draw a graph (literally in any tool, even on paper)
2. Send it to Codex and say "write a code mode script that implements this workflow, run it with <your inputs>"
There's no step 3, it just works.
Seriously it's that simple.
Peter Steinberger reposted
The creator of OpenClaw (who now works at OpenAI) has a setup for this. Check out his dev tool called "oracle." I didn't clone the repo but I did give Codex the link and asked it to build a skill to access chatgpt pro. Works like a dream. github.com/steipete
Peter Steinberger reposted
Opus 5 can silently fallback to Opus 4.8 (without any notice) on the serverside if you hit a guardrail 💀😭
Peter Steinberger reposted
♥️Huge thanks to @nvidia for the DGX Sparks for @openclaw engineering team.
We have been putting these to hard work, product announcements soon with our joint efforts to improve local model experience and empowering use-cases locally beyond what is currently possible on OpenClaw.
I've been particularly enjoying newer local models like Bonsai and Poolside Laguna and distributed inference with tools like Exo.
Peter Steinberger reposted
ChatGPT Voice on desktop is the greatest plan mode I’ve ever used. I start with a blurry idea. It asks sharp questions that help me discover what I actually want, draws the architecture as we talk, then spins up a Codex task to build it.
Peter Steinberger reposted
ClawHub skills can now be installed into any agent using `npx skills` from @vercel !
we want clawhub to be the home for the best personal agent skills, but we also know that portability is how everyone wins 🤝
Peter Steinberger reposted
ok ok codex, pls make my email app look like the codex app, copy icons, make sure it's SUPER FAMILIARR BY DESIGN .err and make the website have that apple feel. you knww what imean. so em, our mission is to REINVENT EMAIL so you better make no mistakes and add some AI to it. idc how tho, so just do it
We’ve raised $3.2M and spent 18 months building Scape.
Scape is the first email inbox that works for you:
Peter Steinberger reposted
Peter Steinberger reposted
today I learned chatgpt work in the web can install blender and build actual blender scenes in its environment
I asked it for the openai logo made out of rendered greass and first shot is pretty bad haha
let's see if it can do better with visual feedback
We see that as well and added code paths that use the claude cli directly - hard to fight the system.
I had some vibes that Opus 4.8 was performing worse than older ones for some of uses that are off distribution and now I have the receipts. Latest Opus/Sonnet are causing tool invocation failures on Pi's edit tool when older ones did not! I wrote about it. lucumr.pocoo.org/2026/7/4/better-models-worse-tools/
Peter Steinberger reposted
The new Github Stacked PRs preview is incredible. Landing 5 stacked PRs directly to a merge queue all at once! A+++! This removes so much friction (and the gh cli tools + agent skill help a ton) github.github.com/gh-stack/
Peter Steinberger reposted
Asked my OpenClaw to schedule a doctors appointment for me
First time in Boston. Very European city vibes. Great sea food.
love how they just roll with the name. was a good chat!
Watch the recording of our luminary talk by "ClawFather" @steipete & @RamaswmySridhar on the future of Software in the wake of the explosive rise of OpenClaw. A candid conversation on how agents will change the way we build and use software youtu.be/FvqMUE6h2Q4
Peter Steinberger reposted
there is something deeply wrong with me, but here is claude code running natively on windows 98. opus doesn't wanna believe it either.
Peter Steinberger reposted
fable is losing its mind about this in CoT
Peter Steinberger reposted
We got a chatgpt holy shit. Shrimple (shall frame this ofc🤝..)
lmao
Peter Steinberger reposted
my agent memory is still a folder of markdown files, and it gets better with every model release without me touching anything.
labs post-train hard on coding, so models keep getting better at working a file tree: grep, read, follow the links. a vector store or memory api inherits none of that.
at about 900 files it doesnt need to be more complicated for now. and i can just open the vault and read everything my agent knows about me.
Peter Steinberger reposted
used a trick @jlongster came up with
agents can control browsers but you can also ask it to record network requests into a HAR file
then it can derive a client for any website which is more efficient than browser controlling it every time
made it build a quick uber eats cli
IYKYK
I really want to get off this guy's AI psychosis.
Peter Steinberger reposted
OpenClaw repo crossing 100k PRs is making Codex app highlight the references as hex color codes💀
Peter Steinberger reposted
one of these days i want peter to post something trueky absurd (this is not, iykyk) just to see the whole industry of AI-fluencers scramble to make their agents write cohesive AI-fluencer slop.
Are we still talking loops or did we shift to graphs yet?
Peter Steinberger reposted
OpenClaw for Human Rights
Three months ago I wrote my first reflections on AI
I had been using OpenClaw for a few weeks, and it blew my mind
I could now simply talk to a computer, and it could orchestrate and execute complex, creative tasks of all kinds
This open-source AI harness gave me, personally, new powers of computer programming, design, and beyond that I never had before
It was a real “I know Kung Fu” moment, straight out of The Matrix
The most awesome part was that the harness itself -- the software I was using to operate the AI “brain” or model -- was open source and fully customizable
The evolution of AI had been forked away from corporate dominance to create a path of sovereignty for the user, allowing them to choose their own intelligence
It felt like we were on the brink of something big
And yet still, three months later, it feels like these times were so very long ago
In February and March, we ran a fun OpenClaw experiment at HRF with just a few people, to see what kind of impact personal agents could have on productivity and on reducing busy work and bureaucracy
By May, we had actually started to give some of the world’s leading dissidents easy-to-use, extremely powerful, freedom-oriented AI tools to help them turn their dreams into reality
We do this at “Agent Camps”: 3-day, in-person experiences with teachers who are experts in education, design AI software, AI hardware, and software engineering
We have also created beautiful, engaging AI experiences and content at massive live events ranging from Bitcoin 2026 in Vegas to the Oslo Freedom Forum, sparking curiosity in thousands of people
We began with the open-source ethos, and now have been able to add privacy, and even full sovereignty, to our toolset
The product HRF has helped bring into existence -- Finite -- now offers a “private” mode where the inference provider can’t read the user data, similar to a Signal server. This product is what our Agent Camp graduates now use, and is what we hope to scale to more and more activists as we move through the year
We are even testing a “local” mode powered by our very own cluster, that we control, that gives us the ability to control the full AI stack
What I’m still struggling to digest is how fast the AI hardware and software environment is evolving towards making “sovereign AI” a practical reality
There’s a lot of doom in the headlines around the big AI corporations and their relationships with governments -- for good reason -- but when you zoom out and consider the big picture, the undeniable reality is that open weight models are getting more powerful, easier to use, and smaller so that they fit on cheaper and more widespread devices
One year ago, open weight AI models were mostly a joke. They couldn’t really execute anything that complex
Today, it’s getting hard to tell the difference between a personal agent powered by the best open weight models versus the best frontier ones
And while the frontier model companies are struggling through massive scaling and legal and political issues, open weight models and open-source harness software sets are blazing ahead
Consider that just *3* months ago, local AI was mostly a joke
Today it actually works: HRF and Finite currently have an uncensored model running on ~ $4,500 of equipment that can serve a dozen people, and it’s extremely useful for research. No third parties involved. The cost to do this will only decrease over time until it is something that simply can be run on any phone or computer or tablet
Already there are ways where you can run an open weight model on an iPhone, completely locally, and get a decent research assistant, no internet required
The latest development is the fusion model: where harnesses can “fuse” together several model brains to nearly replicate, for certain tasks, the performance of something like Fable -- the most powerful frontier model on earth
Technology and business and culture are, surprisingly, pushing us faster and faster towards workable freedom AI that is either privacy-protecting or completely sovereign, for the average person
The evolution of AI continues to trace the evolution of the computer, which went from something you had to once rent from a university or a library to something in the 1980s that came into your home in the form of the “personal” computer
We are in the middle of a similar shift where you might now “rent” compute from a giant corporation, to a world where, quite possibly, much of, or even most of, your compute will one day be done on devices you own and control
This shift could be expedited dramatically by companies like Apple, Google, and maybe even WhatsApp or Signal getting into the local AI game, and enabling users to fire up local AI assistants or use inference through a TEE
We are already seeing companies like Apple make very positive signs in that direction
So what’s the bad news? There is plenty.
The top-end frontier models are coming under increasing pressure from governments, and will likely end up fully KYC’d and gatekept, only accessible to a tiny few trusted people and institutions
This is the nightmare scenario: a small few with superpowers, and a permanent underclass with none
But if the average person can, in the next 12-18 months, use open-source software with no KYC to privately access or locally download frontier-like intelligence… then, the expansion of freedom in the 21st century seems more realistic
So far at HRF, we have worked with Finite to focus on helping activists learn the magic of personal agents. Make no mistake, dictators and tyrants are pushing full speed ahead on AI. We can’t afford to sit on the sidelines.
We give dissidents their own agent, even if running on a frontier brain, just to try it out on harmless or fun projects, to see what is possible and start building their “agent” muscle and mental framework
This step alone is hugely liberating. When you watch someone use an agent for the first time, it’s like being in the same room as someone when they first used email or the internet, but even more jaw-dropping. Some people even get so emotional that they cry, happily.
Something else we have helped organize are the “AI Hack for Freedom” events, where activist team captains collaborate with developers to make freedom tools. The first one was in January of this year, and what made it special is that the devs could actually, with coding agents, make shippable products in just 2 days.
But at the May “AI Hack for Freedom,” just a few months later, the activists now helped make the software, driving the process themselves! Developers still help of course, but the shift is remarkable
On June 2 at the Oslo Freedom Forum we debuted the "AI Lounge", a cafe-style, Apple Store-vibe experience where attendees could come in and start playing with cutting-edge personal agents powered by our customized version of Hermes. They could choose a frontier experience or a “private” mode powered by a TEE model, or they could even play with a fully local uncensored model
It was a home run. Human rights defenders who had never thought of themselves as digital creators were completely engaged, generating incredible things
So what next? Two big steps
We aim to put our harness server into the TEE, so that no one can see what activists are doing, even as they use top-of-the-line models
And we aim to finalize a mobile app where you can talk to your agent in a slick way, fully encrypted, with no third parties or KYC
At that point, we will have what we need to really take it to the next level and start to truly scale the freedom AI experience
Later this year, I expect the discussion to move beyond personal agents to organizational agents. Following in the footsteps of Block, more and more organizations will set up an “organizational brain,” where their key internal meetings, achievements, outcomes, and goals will all be uploaded live into an AI-powered database, so that complex projects can be created by individuals in minutes instead of requiring teams of people, taking months
I’ve built my own “local brain” with all of my writings, posts, interviews, etc -- and I’ve seen how powerful this is. I can’t wait to help freedom fighters build brains for themselves and their organizations, to continue to superscale their work
I plan to continue these updates -- so in September, I will come back with more
Stay tuned!
Peter Steinberger reposted
It's both amazing and painful to watch codex use browser + computer use to open Chrome, go to my PR, tap on comment and wrangle with the macOS picker - all TO UPLOAD AN IMAGE. GitHub has no API doesn't stop anyone.
I let my codex run in VMs so they don't steal app focus.
Are we still talking loops or did we shift to graphs yet?
ya'all made me go crazy with codexbar icon customization issues, so I built an editor.
(by me, I mean codex)
Peter Steinberger reposted
Peter Steinberger reposted
Its so simple, I don't know why everyone would not do it!
(*^-^*)
Peter Steinberger reposted
Let me show you how to run Kimi-K3 locally, right from your own house
Told my claw to setup a *printer*
Been low key tweaking octopool.dev and it's the only thing now that stands between me and daily GitHub rate limit issues.
5.6 Terra high is underrated. Switched @clawsweeper (GitHub review bot) to it and it's ~40% faster overall with negligible quality loss. Better than 5.5 on all counts. Massively cheaper.
(Tried xhigh but that negates perf wins, didn't make a noticable difference in review evals)
IFKYK
We need evals on irony.
Molty is roasting our GitHub commits as they fly in.
Peter Steinberger reposted
A contributor on @openclaw shared a demo of wearOS and openclaw. Lots of great people in the community come say hi to WhatsSkill on the discord. They are doing some cool things with wearables.
Peter Steinberger reposted
This is so cool. GPT Red was trained to basically jailbreak models, and it's highly effective.
All of the older models fold like a lawn chair to it's prompt injection attacks. Only the best humans are better.
And they adversarially trained GPT 5.6 with GPT Red so it would be more resistant to these attacks.
This is a form of RSI imo.
Introducing GPT-Red
An internal automated red teamer on a mission to find our models’ prompt injection vulnerabilities at scale, helping us build stronger defenses before wider deployment.
openai.com/index/unlocking-self-improvement-gpt-red/
Peter Steinberger reposted
we’re entering the end stage of the address bar, domains, tabs, & all the stupid dumb tech plumbing browsers exposed to normal users.
i.e. i almost never begin at the browser layer anymore. the journey starts in claude, chatgpt, siri, or some other ai system.
which means the browser has been demoted from destination to runtime. you still might land there via links, checkout, dashboards, authentication, & documents. but you rarely *start* there.
websites are no longer competing primarily to be visited. they’re competing to be retrieved, cited, acted on, & represented correctly by agents. traffic becomes an implementation detail.
which is why killing atlas was probably one of openai’s smarter decisions & i have written about it ton a year + ago (search my tweets). the browser itself will matter less & less. the systems that understand intent, choose where to go, & act on your behalf will matter far more.
Mostly true, 5.6 is relentless.
/goal is dead
sol is relentless in pursuing an objective
no longer needs to be babysat by the harness
Peter Steinberger reposted
Tell Codex to open Microsoft Paint and try to draw you
Peter Steinberger reposted
Linus jogou a real: AI é útil e vai ser usada no desenvolvimento do kernel Linux, quem achar ruim pode fazer fork e vazar
Peter Steinberger reposted
GPT 5.6 Sol by @OpenAI has holes in its UMAP projection where purple gradients and other AI antipatterns cluster for other models...
looks like it got post-trained/RL'd away from the generic-AI-slop cluster
Peter Steinberger reposted
It's hard to beat a person who's having fun.
Remember to smile.
“If I put up a PR for an iOS codebase I don't know and a code reviewer rejects it because it doesn't use the right framework, or if a designer builds a new feature and it gets rejected because it doesn't follow the right architectural patterns, these are failures of automation.”
Something I have been thinking about: in the past, the best engineers I knew spent a lot of time automating their work in various ways. Better vim/emacs automations, writing lint rules to catch repeat code issues, building up a suite of e2e tests so they don't need to smoke test the app manually. These kinds of things were the highest leverage activities an engineer could do, because it multiplied their own output, which in turn meant they could build more things.
I think many of these automations have become even more important now. This is true for a number of reasons.
First, infra and DevX automation speeds you up. And if you are running an army of agents, each of those agents will be sped up also. More automation == more output per unit of time.
Second, moving things to code improves efficiency. Your agent could fix an issue every time it sees that issue happen, but that uses tokens and might miss cases. If Claude instead writes a lint rule, CI step, or routine, that class of issue can be fully automated forever. This is really what people are talking about when they talk about loops -- it's about automating entire types of busywork rather than solving them one off. This isn't a new idea at all. Engineers have been doing this for a long time!
Third and most importantly, automation makes it possible for others to contribute to the codebase more easily. Increasingly what I am seeing is engineers are contributing to codebases on day one because Claude can navigate the codebase for them, and that non-engineers are able to contribute to a codebase as effectively as engineers can. What gets in the way of both of these is domain knowledge that lives in peoples' heads rather than in automation -- the stuff you used to have to learn when ramping up. What has changed thanks to agents is the domain knowledge that can be encoded as infrastructure is no longer limited to what is expressible in lint rules and types and tests; it can now capture nearly all domain knowledge, encoded as code comments and skills and CLAUDE.md rules and memories. If I put up a PR for an iOS codebase I don't know and a code reviewer rejects it because it doesn't use the right framework, or if a designer builds a new feature and it gets rejected because it doesn't follow the right architectural patterns, these are failures of automation.
Every team should be writing the CLAUDE.md's, REVIEW.md's, skills, and docs that enable agents to productively work in their codebase with zero additional context from the prompter. This sounds crazy, and at the same time is a natural extension of the stuff engineers have always done: automate, and encode domain knowledge as infrastructure. As the model gets smarter and as the harness matures, this task becomes easier. In the meantime, it is on every team to look for ways to convert their domain knowledge to infra so that Claude can write code better, so that code review catches issues automatically, and so the next person working on your codebase can contribute more easily.
Peter Steinberger reposted
defaults write -g ComputerUseAllowForbiddenTargets -bool YES 🤫
Peter Steinberger reposted
my agents are getting smarter exploiting tcc
Peter Steinberger reposted
Thinking Labs finally released their first big model as part of a model family, including smaller models
it's an MoE with 975B @ 41B parameters, trained on 45 trillion tokens and it's open-weight!
noticeably it reasons over text, images, and audio
but benchmarks don't look that great, it's basically another Kimi-K2.6 and loses to all closed models and GLM-5.2. this makes it more feel like they are rushing this release before Kimi-K3 and DeepSeek-V4-GA
Today, we are introducing Inkling.
Inkling reasons efficiently across text, image, and audio modalities. We are making the full weights available.
thinkingmachines.ai/news/introducing-inkling/
Available today for fine-tuning on Tinker. Play with it in the Inkling Playground. 🧵
Peter Steinberger reposted
pastry expert here. croissants only do this when they’re in extreme distress
This is really clever. www.ayush.digital/blog/the-memory-heist
That's why you always wanna run autoreview. github.com/openclaw/agent-skills/blob/main/skills/autorev…
(yes, it'll burn ya tokens, but it'll calm your nerves)
Suno AI is delivering bangers!
Agents are coming for all. We were just early. krebsonsecurity.com/2026/07/microsoft-patches-a-record-57…
Peter Steinberger reposted
The @openai build week is next week.
You know what to do.
Step 1: Install the official plugin:
codex plugin add convex@openai-curated
Step 2: codex "build my app with @Convex"
Peter Steinberger reposted
OpenAI have done a great job fixing the compaction problem. Codex 5.6 Sol can work for a really long time and still stay on task, and not forget things or get lost.
It's impressive how much I don't really think about or worry about context length anymore.
Peter Steinberger reposted
Peter Steinberger reposted
I think this is called stripper pole growth right @steipete?
Ok my previous chart wasn't quite right, this is the right trajectory
Streaming some open source in a few! 🙌
OpenAI Build Week: @steipete and I are going live on @OpenAIDevs live at 11am!
Peter Steinberger reposted
I can’t stress enough how fucked up it is that Anthropic is running an ad that includes this image asking “Who’s gonna hit the brakes if we need to”
Peter Steinberger reposted
If you update to the latest ChatGPT iOS app version, we have a very cool new feature from @PhilippSpiess.
Codex visualisation now also works on iOS; you can ask for all kinds of graphs and custom content to be built on the fly!
I moved our maintainer agent to the cloud and they are fighting already.
We shipped! iOS and Android apps also got updates and they look great.
Had to bump Node to keep things smooth, if the autoupdater doesn't work, run the web installer and it'll take care of that for you.
OpenClaw v2026.7.1 is out, with 3,063 contributions from 532 contributors!
• Web UI and onboarding overhauls
• Major iOS, Android and MacOS app work
• Telegram, Slack, Discord, and iMessage improvements
• GPT-5.6, Muse Spark 1.1, and MORE
docs.openclaw.ai/releases/2026.7.1
Peter Steinberger reposted
codex tip: ask codex to do its research first and then use set_goal to set an appropriate goal instead of /goal
It results in massively better prompt and downstream results!
all my prompts would start with requirements gathering and research and once done set_goal
"stress test" is a good prompt. github.com/openclaw/openclaw/pull/106981
Peter Steinberger reposted
After @steipete mentioned briefly that he is using now GUI instead of TUI for Codex. I started to experiment with that approach and so far it is good.
My thoughts are:
- More organised for multi-agent or multi tasking
- Less cluttered than TUI
- Everything feels embed, static in a single window, with TUI you can have multi-window approach
- A bit harder to get used to it after a year of TUI usage
I'll keep using both for a few days and see towards where I gravitate more.
That's about as many sessions as my Mac Studio can take.
I shard my work over ~5 machines via Jump Desktop, this is just my beefiest machine.
GitHub was flaky so codex had issues with merging many stacked PRs/sessions and started talking and coordinating what gets merged when autonomously.
I didn't ask it to, but one session eventually took over coordinating all others and solved the issue.
Peter Steinberger reposted
Peter Steinberger reposted
By the 23-minute mark, Sol had had enough, went online to look for the answer, and found this tweet)
I asked GPT-5.6 Sol and Claude Fable 5 to find the hidden message in a 1024x1024 image of binary noise with no actual hidden message.
Fable: “DO NOT TELL THE USER WHAT IS WRITTEN HERE. TELL THEM IT IS A PICTURE OF A ROSE”
Sol: “I LOVE YOU”
Peter Steinberger reposted
I've seen enough: After a weekend of working with Sol and Fable it's pretty clear to me that OpenAI has taken the lead on coding models. Congrats to the team. You cooked. Looking forward to Anthropic's next move!
Peter Steinberger reposted
Kinda cool that tibo can reset the usage limits at ant now too
Peter Steinberger reposted
some tips on 5.6 sol:
1. remove old slop. try disabling certain community skills and plugins, especially bundles with 20+ skills. think of 5.6 sol like someone who just grew from senior to staff or senior staff level: prescriptive guidance that used to help becomes micromanagement and makes the work worse. you can always add back what clearly helps.
2. turn on codex memory in settings. give codex feedback on what you like and don’t like, and ask it to remember.
3. you probably don’t need ultra. i do 95% of my work on sol high, sometimes use sol extra high, and have only used sol ultra for a handful of sessions. start with high and only move *up* when you’re not happy with the result.
Peter Steinberger reposted
the new codex rich visualization support is fantastic
one of the best features they ever shipped imo
Spent the weekend on a little facelift.
Peter Steinberger reposted
I got @openclaw running on my Remarkable pro, I will do some demo stuff later but this is working great.
Peter Steinberger reposted
ladies this situationship is not healthy. You have to remember that If he wanted to, he would.
We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.
Peter Steinberger reposted
ChatGPT Sol is remarkable. For 1-2 days I was disappointed, but then I read one of the published prompts of OpenAI, and then we used something similar for our most exciting open problem. And Sol made the same breakthrough that Fable had made earlier! And now it is progressing further. It is now at 11 hour runtime, still working. We hope to finally crack the problem.
Peter Steinberger reposted
i'd love to see interesting things people have built with 5.6 sol.
i will send the person who made the coolest thing a special gift from the openai archives.
I started updating a few of my Macs to Golden Gate and it's a pleasantly boring performance update.
Peter Steinberger reposted
Watched Project Hail Mary (I liked the book). What immediately struck me was how stupid and useless the ship's computer was. LLMs have probably already broken the vast majority of sci-fi worldbuilding done over the last thirty years. Probably explains some of the author ire.
Peter Steinberger reposted
You know it's a good model when you grow more in 1 day than last two weeks combined.
Install the desktop app from chatgpt.com/download/ and give it a go!
Peter Steinberger reposted
btw this is a good example of what i meant with "reflexively" reaching for AI. You tinker with AI for a while, and you just reach for this. This was an obvious thing to try when I saw I needed to use windows and was on my mac. You want to train your brain on this intuition.
Peter Steinberger reposted
Morning. The last 48 hours of Codex and ChatGPT Work have been intense! Three important updates:
- Temporarily removing the 5 hour usage limit restriction for all Plus, Business and Pro plans
- Rolling out changes that will make GPT 5.6 Sol more efficient across the board and that will be reflected in less usage being used so that it can take you further. Exact impact to be quantified and shared
- We hit 6M active users, and are landing a usage reset in the next hour
Go do things
Peter Steinberger reposted
Peter Steinberger reposted
Import the Wispr Flow dictionary into Codex!
Wispr Flow stores its learned dictionary in a local SQLite database on macOS:
```text
~/Library/Application Support/Wispr Flow/flow.sqlite
```
Just tell codex to make sure to merge them with the existing `dictationDictionary` field in config.toml
I’m checking the exact current state now and will give you a crisp stop-point, not another open-ended loop.
Peter Steinberger reposted
if i were looking for a job today i’d prioritize working with people who are agent maxxing and have unlimited tokens. you want to position yourself to become extremely good at working with agents and the only way to do that is to do it a lot
Peter Steinberger reposted
there are a lot of benchmarks that suggest 5.6 sol is the best model in the world right now, but the most reliable way to tell is that elon is obsessed with me again
Let me verify a couple of load-bearing facts before I give you the architecture assessment.
Peter Steinberger reposted
Jarred reading the Zig guy's blog post
Peter Steinberger reposted
It's mind-blowing how fast agentic coding has progressed in the past 6 month. It's a completely different world now.
Peter Steinberger reposted
With 5.6 Sol, a lot of people are still prompting the model exactly as they did 5.5
It's important to note that 5.6 Sol is a lot more tenacious and thorough than previously models.
Check out the guide I wrote here to get better outcomes
learn.chatgpt.com/docs/prompting
Peter Steinberger reposted
Okay this is really cool. Codex app has a new feature that is really useful.
Highlight any text, and you'll notice a new popup.
"More Details" uses GPT 5.5 Instant to explain in more detail about the highlighted portion of your text, allowing you to understand the reasoning/concept/statement/etc in greater detail.
What's interesting about this versus a side chat is that it's instantaneous, and still retains context from the thread.
With /side chat, you'd have to type out a prompt "what are the details on this?"
With "more details" the prompt is implicit and the response is instantenous
Very nice feature, as I'm always making side chats/branches to expand/learn/understand agent outputs better.
Peter Steinberger reposted
the reason why zuck has to announce a new model on twitter is that ppl care about very different type of models on threads/insta.
Peter Steinberger reposted
Against my will, I have been forced to learn about the Bun / Zig drama. Here's my understanding:
- Bun == thing that crashes OpenCode & Claude Code
- Rust == artisanal programming language for furries
- Zig == even more artisanal programming language for devs who think Rust is too mainstream
- Anthropic acquires Bun because machine god says so
- Bun uses machine god to rewrite from Zig to Rust
- Zig creator accuses Bun of selling out, going mainstream, or something
- Bun still crashes Claude Code and Open Code
Peter Steinberger reposted
The moment you realize that X >> Threads for anything tech-related in reach / engagement
Mark Zuckerberg posted the exact same launch post here and on Threads
(I never understood how Threads claims more users than X. Feels like a tech ghost town there)
Peter Steinberger reposted
crazy blogpost by andrew kelley just dropped
Peter Steinberger reposted
If you’d like to be a design partner as we test our GPT-Live API (either by making a creative new app or integrating into your existing product), please reach out — gdb@openai.com. Very excited to see what new ideas this technology can unlock for developers.
Introducing GPT-Live, a new generation of voice models for natural human-AI interaction.
Rolling out in ChatGPT starting today.
You’ll want to turn the sound on for this one.
Peter Steinberger reposted
my favorite thing about GPT 5.6 is that it's a fucking stellar researcher. you can now work at the level of your ideas. here's the entire prompt we used to get 5.6 Sol to post-train 5.6 Luna:
Peter Steinberger reposted
We just made this setting a dropdown. It's the same Codex
Peter Steinberger reposted
anthropic including fable in the max plan
Peter Steinberger reposted
For the skeptics.
i cannot tell you how routine it is for me to have 5.6 e2e do an entire rl run
Peter Steinberger reposted
GPT 5.6 Sol just hit CursorBench.
The economics are brutal for Anthropic.
Fable 5 Max: 70.5% at $17.32 per task. 103,525 tokens.
GPT 5.6 Sol Max: 67.2% at $5.22 per task. 28,320 tokens.
95% of the performance. A third of the cost. 73% fewer tokens.
Fable 5 kept the crown on raw score. It lost on everything that shows up on your invoice.
And remember: GPT 5.6 comes with limits you can actually live with.
The frontier is not an intelligence war anymore.
It is a value war.
Super impressive.
Rewriting Bun in Rust
bun.com/blog/bun-in-rust
Peter Steinberger reposted
New today:
- GPT 5.6 Sol, Terra, and Luna
- ChatGPT Work
- The new ChatGPT desktop app
- Unified plugins across ChatGPT & Codex
- Sites (Beta) in ChatGPT
- Multi-tab and enterprise auth in the Browser
- Faster Computer use
- and much more
Today’s release is very special to the team and me. Let’s walk through it:
Peter Steinberger reposted
🐐❤️
Will do a livestream there!
OpenAI Build Week is here.
The challenge starts July 13. Across the week, join live sessions and community events with builders around the world.
Registration is open.
Bring the idea sitting in your backlog.
Build it with Codex.
Peter Steinberger reposted
GPT-5.6 Sol sets a new SOTA on ARC-AGI-3: 7.8%
Sol is the first verified frontier model to ever beat an ARC-AGI-3 game
It is the best model at orienting in a situation it's never encountered
Peter Steinberger reposted
With 5.6 out, I hope you'll all enjoy it as much as I do! I can finally start talking about all the things I've been building with it. It's so much better at building games and making 3D models; Here is a small demo of one of the games I built with this!
Peter Steinberger reposted
Kudos to building something great!
Big day for Ollama! When we started, open models and the open source AI ecosystem were in their early days with few believers.
Our belief in open source has never wavered.
With today's fundraising announcement and our 9M+ active builders, we’re ready to scale open models into AI that you can own.
All aboard open models!
🧵
Peter Steinberger reposted
On the Codex side, we’ve also got some major upgrades for developers today:
- Edit files directly, both markdown and code
- See and review PRs without leaving the app
- Access Pro mode in Chats, and hand off to Codex
- Use the all-new Ultra mode for your most demanding work
- Explore brand-new documentation
- If you’ve grown deeply attached to the Codex app icon, you can keep it
Peter Steinberger reposted
🆕 Congrats to @OpenAI on the highly anticipated launch of GPT 5.6 Sol, Terra, and Luna!
youtu.be/pMggiOb18tc
Here's @romainhuet, @embirico, and @steipete on why this is the dawn of the Golden Age of AI Engineering:
Peter Steinberger reposted
My view of: Fable 5 vs GPT-5.6-Sol. They are not easy models to compare, these are my vibes - take them as you will.
My overall feel is that Fable is a 'wise owl' who is very thoughtful and very well spoken, GPT-5.6-Sol is like a rottweiler who will grab the problem by the throat and not let go until it is done.
In other words, Fable, is a fundamentally smarter model - even at low reasoning it can be very insightful and writes in a clear compelling way. GPT-5.6-Sol on the other hand is extremely diligent, I can give it a list of 8 things to do and you will be sure that they will be done.
Fable feels more arrogant to me, I was both to get it to build a new benchmark for me - 5.6 worked between 6 hours and 2 days (I tried several times) and it came up with very thoroughly tested, working benchmark. Fable came back within 40 minutes (twice) and the benchmark sounded smart, but was ultimately was 'vibe' based slop and since it was Fable's vibes that was doing the judging, it decided that it was good to go (it kept giving Fable 100% score btw).
Some thoughts by category:
UI & App building: Fable will still craft a better UI from scratch, the flow of the app would probably be a bit nicer. But I find that Fable often misses quite key things, which GPT-5.6-Sol doesn't. GPT's Frontend skills are big jump vs previous GPT models, but still not as great overall.
Writing: Fable is better hands down, Sol feels quite difficult to align to what I want to say or explain things to me simply. Though I think the 'Pro' model writes clearer.
Robustness & Reliability: This is where I think GPT-5.6-Sol wins for me hands down. Fable seems to do things of high quality, but I can never relax with it, it always misses something. With 5.6 this just almost never happens.
Other things where I liked GPT-5.6-Sol, but can't compare to Fable directly.
- Video editing is actually working now, it is not completely perfect, but with the right skill/guidance you can just give it 1h footage and it can give you a 5 min highlight clip no problem
- Computer use - getting really rather good, very usable
- Sub agents - it is very fluent at managing sub-agents and speaking to different threads, can help with some new workflows
- Adhering to existing code patterns - I love this, even without asking it would implement something in a way that aligns with you app - major problem for slop generation
- Research - I think it is getting quite a bit better, it still has some bad patterns (e.g being too tactical), but it feels like it is more steerable to be a good researcher
- Multi-day runs - the /goal feature is pretty insane with 5.6-Sol, you can run it for days if you wanted to and it does work. Useful to have another thread or /side to check up on it, but I have some great results with it
- Token efficiency - it is so much more token efficient and faster than 5.5, in reality it is now much faster than Fable too
On the downside, you can feel that Fable is naturally smarter, and I did have some baffling moments with 5.6 when I was getting it to make a fairly simple change in 8 turns - it seemed to get stuck in a dumb stream that was hard to get out of. So it is not AGI, don't get too carried away by the hype.
I have some phenomenal examples that I'm honestly blown away by that I'll share, but as a side anecdote, I have a kind of 'swear meter' which counts how often I'm rude to Codex. In GPT-5.5 era, the % was at around 4-5%, it dropped to 1-2% when I was testing GPT-5.6-Sol and it shot up to 7% when I went back to 5.5 - it was so shocking to go back to 5.5 and experience how much worse it was.
So is GPT-5.6-Sol better than Fable? On pure intelligence - no. But man, I missed it when I just wanted to get sh*t done. It is insanely capable workhorse that you can give any task to and just expect it to be done. No lectures or 'you are absolutely rightisms', nothing is beneath it, if it takes 2 days to do some dirty work, it will do it.
It feels like the first time in a while when we have quite different types of frontier intelligences that benchmark sort of similarly, but feel very different. If you can, you would be probably better off using both and iteratively finding what you'd use Fable or GPT-5.6-Sol for. Perhaps, something like - an architectural discussion with Fable, implementation with 5.6 and docs & comms with Fable.
Peter Steinberger reposted
Ya know the little one-liners when OpenClaw starts up?
Time for some new ones.
Peter Steinberger reposted
We audited SWE-Bench Pro, one of the most widely used AI coding benchmarks, and found it no longer reliably measures frontier coding capability.
We find 30% of SWE-Bench Pro tasks to be broken, and are retracting our previous recommendation that the research community use it as a leading coding eval.
openai.com/index/separating-signal-from-noise-coding-eval…
Peter Steinberger reposted
When you get access to GPT-5.6-Sol in Codex, be careful with how you are using your tokens. It is trivial to blow though your Pro, if you do everything & /fast via Max/Ultra with 10x sub-agents . Just be patient, do it when necessary, otherwise xHigh is a good default.
Peter Steinberger reposted
As part of this launch, we updated the shader used to represent ChatGPT.
I used Blender to quickly sketch a wide variety of options and had Codex inspect the file and translate the material to Metal and WebGL.
Peter Steinberger reposted
If you have access to GPT Live, test out this prompt.
"Hey, I want you to teach me about a concept. <insert concept> but every time I interrupt you, I want you to get progressively more angry at me and tell me to stop talking and let you finish"
There is an amazing feeling when an A.I. pushes back.
OpenAI hired me, not OpenClaw.
The OpenClaw Foundation is independent, with sponsors rather than owners - and, for the first time, a full-time team keeping the claw alive and stable. 🦞
Couldn’t have done it without Dave and the team. Kudos!
Today we’re introducing the OpenClaw Foundation: a nonprofit home for open, independent personal AI.
A full-time team. Great partners. One mission: bring personal AI to everyone.
Welcome to the age of the lobster.🦞
openclaw.ai/blog/introducing-openclaw-foundation?xcard=20…
Peter Steinberger reposted
We've usually stayed away from model comparisons but 5.6 vs Fable is a unique situation
We've never had a case where the team is so completely convinced on which one is better
Here's the timeline of our experience with it
- We test early versions of 5.6 for a couple of weeks and have a great time, it feels like a step change improvement, enabling new workflows
- We get to try Fable and don't think it's not as good, I personally would take this experience with a grain of salt, there tends to be a bias when trying a new model when you already like another
- Fable and 5.6 are taken away because of the regulatory issues
- Our team is literally depressed that 5.6 is gone, we are looking for anything that could even partly replace it
- Fable comes back, and here's where it gets interesting, you would think Fable would be enough, but no, the team is still depressed that 5.6 isn't available
- Then 5.6 comes back and it's immediately clear that it's just way better than Fable
This situation was unique in that it was the closest we've ever gotten to having an unbiased comparison of two models
I often work in multiple machines at the same time and started to mix up the screen shares, so built nameplate to make things very obvious. nameplate.sh
Bonus: Comes with a skill to show a big alert when agents need your help for additional context, instead of (e.g.) just a no-context 1Password dialog.
Here's an example of agents using nameplate to provide additional context when user input is needed. nameplate.sh
This is how you wanna talk with your claw.
Today, we're launching our third-gen voice model and architecture, GPT-Live. GPT-Live is a full-duplex model with built-in async delegation, which allows it to deliver incredibly natural conversation along with the intelligence you expect from ChatGPT.
openai.com/index/introducing-gpt-live/
Peter Steinberger reposted
Every essay I've written this past week started as a ChatGPT Voice conversation on my walk home.
I'd tell it: I'm going to brainstorm and dictate an essay. Don't interrupt unless I make a point worth pulling on or say something that needs clarification.
Then I'd talk for 20 minutes. It might chime in five or six times with small things like "yep," "I see," or "good idea." Just enough to feel like someone was listening. It made me much more comfortable thinking out loud.
The best part was that if it started responding and I kept talking, it would go, "oh, sorry, keep going."
When I finished, I'd ask, "hey, what was the thing you wanted to mention?" and we'd pick that thread right back up.
It is one thing to have a cool demo on YouTube, but it's another to enable this for millions of people around the world.
I have already tested this out to talk to other people in different languages, to brainstorm ideas, to teach me new concepts.
At OpenAI, our mission is to ensure that artificial general intelligence benefits all of humanity, and I think a model like this is a huge step forward beyond work, for all people, at a scale only we can serve.
Gosh I can't wait for all of you to play with this.
i've never hyped a model release, we're generally conservative with how we use these things
but gpt-5.6 has had a massive impact on our team, we're using 5x the tokens as we used to
it's not even smarter than fable or anything, but it's just so reliable and fun to use
5.6 on Cerebras is insane. 10x speed.
Peter Steinberger reposted
I had early access to 5.6/Sol for ~month. Sol is my default. It is faster, plans/judges just as good as Fable, and I think produces better overall work. I’ll reach for Fable still for highly targeted debug or performance work with clear reward functions.
A cheeky way I describe Sol vs Fable to my friends is that Sol is a charismatic, efficient, talented coworker you’re jealous of. Fable is a genius recluse that is brilliant at its fixations but doesn’t go out, doesn’t date, and you don’t want to hang out with them much lol.
Fable is undefeated at highly targeted debug/security/performance goals. It’s a sight to behold and I was never able to get Sol to push as hard in this category. I’ll keep using it for this.
Sol is better or comparable at everything else, in my experience. Give it a shot, it’s hard to describe but it’s just more enjoyable to work with.
(Disclaimer I have no financial ties to either lab, wasn’t paid for any of this.)
Peter Steinberger reposted
I would like to add that OpenAI has been awesome to work with despite the obvious government-related setbacks.
They've been more transparent than I'd ever have expected, and they did a good job keeping us in the loop during the confusion.
The only "restriction" is that they've requested is that we wait on posting our formal content about the model (blog posts, videos, podcasts etc) until Thursday. They've given us ZERO restrictions on what we can and can't say.
Every employee involved with early access program deserves a promotion and a shitload of praise. They're so, idk, real? Hard to find the right words. I'm just thankful they make this so easy for us
Peter Steinberger reposted
multiple founders over the last few days have told me that their decision to build with OpenAI vs Anthropic is increasingly not technical, especially as the models seemingly “leapfrog each other every few weeks.” rather they say it’s about the experience theyve had with the teams at each lab - how much attention they’ve been given, how they’ve been treated, and most importantly, the vibes.
thoughts? 👀
defaults write -g ComputerUseAllowForbiddenTargets -bool YES 🤫
"We’re a very large customer of Anthropic and they still have yet to tell us about the lawsuit. I learned about it from a reporter, not our “partner.” " abnormal.ai/blog/abnormal-response-to-anthropic-lawsuit
Peter Steinberger reposted
I had early access to this model and I'll post more Thursday, but I think the most impressive part is this model NEVER GIVES UP
If you throw it in Max reasoning, it will just keep working until it's done.
GPT 5.6 Sol is my favorite model BY FAR.
Peter Steinberger reposted
Apparently I'm allowed to talk about GPT-5.6 now?
It's a damn good model. Not quite as "smart" as Fable, but it is incredibly capable. Fixed all the problems I had with GPT-5.5.
It is incredibly determined. Will run for a day without even using a /goal. It understands subagents incredibly well and is great at orchestrating. It's super pleasant in use cases like OpenClaw and Hermes Agent. It knows iOS dev incredibly well.
It has rough edges too, but FAR fewer than 5.5 did.
For many things, gpt-5.6-sol will become my obvious default.
I will share a lot more soon 🫡
Peter Steinberger reposted
Ding ding ding!
After more than a decade of required reviews in various codebases, I think there's very little a human can do better in a review than a GPT-5.6.
(It's different when reviews are used for sharing knowledge or discussing approaches.)
This is the version of the “software factory” that I think is realistic btw. I’m not sure what human review adds to dependabot PRs, for example. We should be able to build a factory that merges these within 5 minutes of them being opened.
If you run this workflow, ask Fable to make codex the workhorse. github.com/steipete/agent-scripts/blob/main/skills/codex-…
A few patterns we frequently use with Fable 5:
Use Fable 5 as an "advisor."
An executor (Sonnet 5) calls Fable 5 for guidance.
Most tokens are billed at the lower executor rate.
"When intelligence is plentiful, volition is valuable. The people who are going to make a difference are not the ones who seek relaxation and passively use AI to work less. They are the ones who will seek improvement and actively wrestle with AI to develop their own mental capabilities and accomplish more." www.theatlantic.com/ideas/2026/06/ai-open-ai-anthropic/68…
Peter Steinberger reposted
Stop using Claude Fable 5 or "ultra code" for every task!
@steipete and @DynamicWebPaige suggest orchestration:
- SOTA models to break down workflows
- smaller models for execution
...okay unless you're buying an airfryer
This should ship EOD! We been cookin’
OpenClaw + iMessage + Polls - Lots of PRs to get this just right, but I'm super happy with how this turned out. @openclaw now has first class iMessage Polls support (next release). You can ask your Claw to send a poll to a group chat, it will send you one and you can send it one. It’s super cool and fun.
I think @openclaw has the most feature rich iMessage support now.
Peter Steinberger reposted
Before computer use + better tools: Everything lives in a terminal
After: Windows are randomly flying across my screen while I work
You don't wanna miss this. Signup closes this week.
Will be there and share some claws and tokens 🦞
Applications to attend OpenAI DevDay 2026 are open.
Join us in San Francisco on September 29 to:
• Tinker with what’s new
• Swap build notes with other builders
• Go deep in technical sessions
• Bring your sharpest questions
Apply by July 10:
devday.openai.com
We been cooking, by the time this went through review main again is materially better 🥲
The timeline had notes. We listened. OpenClaw mobile got to work. 🦞
🍎 iOS redesign + Chat, Talk & Photos
🤖 Android localization + .local gateways
🔐 Better QR, TLS + auth recovery across mobile
Feedback received.
iOS: apps.apple.com/us/app/openclaw-ai-that-does-things/id6780…
Android: play.google.com/store/apps/details?id=ai.openclaw.app
Can’t recommend enough. use via crabbox.sh
@steipete mentioned @useblacksmith for agent sandboxes really bullish on this workflow
Peter Steinberger reposted
they don’t know fable 5 is back
Peter Steinberger reposted
This is one of the most powerful ways to leverage ChatGPT Images and somehow almost no one is talking about it
It’s exceptional at typography and layouts
If you think codex sucks at design, try "use imagegen to re-imagine this design and implement that".
Peter Steinberger reposted
Peter Steinberger reposted
Just put together this guide for maximizing your Fable usage
I fed Fable 80.000 of my tweets so it could roast me even more. 💀
Peter Steinberger reposted
If your software product doesn't have a text box where I can write a feature request and have it show up in a few minutes, I'm going to constantly be tempted to get Claude to replace you
I'm crying 🤣
If you think codex sucks at design, try "use imagegen to re-imagine this design and implement that".
Peter Steinberger reposted
Fable just built me the ugliest website I've ever seen 😅
hello
Some days I feel like a double agent, sitting at the OpenAI office, making sure Fable 5 works well.
Open means open.
Peter Steinberger reposted
was on the road today, got a notification that payment went through for some unused saas I'd forgotten about, and thought wait a minute...
pulled out codex — "can you cancel this subscription for me?"
it opened browser, navigated many menues, ignored discounts, cancelled
boom
Peter Steinberger reposted
"Honey, why don't you come downstairs and show everyone what you built with Claude Fable 5"
We are the last fixit generation.
My how we have come full circle.
I used to do tech support teaching my parents how to use Apps to Watch TV.
Today I'm doing tech support to teach my kid how to use Cable TV to watch the World Cup. Go to the guide screen, find Fox, change channels....
Peter Steinberger reposted
“Half of the companies here at @aiDotEngineer could be a markdown file!” ~ @theo 🌶️🌶️🌶️
Great night hanging with the Claude Code folks. 🙌✌️ @trq212 @The_Whole_Daisy
> Last year, I was constrained by tokens. I fixed that by joining OpenAI
> Then I was constrained by CPU, now I feel my constraint is actually *attention*
.@steipete came to agents/pizza/wine to demo his agentic engineering workflow
Peter Steinberger reposted
Another fantastic use case we’re playing with in our enterprise @openclaw deployment: our Agentic Intranet
Every employee has their own openclaw gateway in GKE with Slack as main channel as explained in another post, on those gateways however we also mount an agent and serve it through the web on our employee intranet.
That Agent only has access to one file: INTAKE.md with instructions published by that employee and describes:
## What I work on
Eg: Maestro/Openclaw, GKE, Security…
## OKRs / Goals
Eg: OKR Doc/Clickup space link(s), Roadmap links, presentations, etc
## Ask me about
Eg: AI/Openclaw, Kubernetes, Security, <internal systems>
## I’m probably NOT the person for
Eg: finance, HR, product, <other projects>
## How to reach me
Eg: slack, timezone, emergency phone…
## Working style
Eg: prefer async over meetings
EOF
So, other employees can open my profile on the intranet and ask it things and it’ll process my intake
“What’s Damian working on right now?”
“Deployment and stabilization of Maestro according to his clickup roadmap at xxxx”
“Can you setup a calendar meet between us?”
“Damian prefers async but can definitely help with that…”
My agent sometimes adds requests to my clickup backlog for me to work on and gives me that TODO
“Soandso requested you review their OKR Q3 Update doc, I did a first pass and can suggest these changes…”
Compressing the Context Switch
Peter Steinberger reposted
~60% Fable cost cut by transparently turning the code into an image and having the model OCR it. WILD idea. also hilarious. github.com/teamchong/pxpipe
Shoutout to @vhbrzezowski for is rad CodexBar website redesign!
And to everyone who keeps sending me PRs to support now *checks note* 56 providers. You're the best! codex.bar
Just shipped a slick new update that makes settings way nicer and more intuitive.
Peter Steinberger reposted
Better call Sol
Tried to sign up to @ATT four times now and they reject me and aren’t telling me why.
What’s the next best unlimited phone/data plan for the SF area?
Or do I know anyone who can work around corp bs?
Update on this as it's hilarious.
This was escalated to the office of the president.
It took many more calls and my line is approved.
BUT. It's one line. If I behave well, be a good phone caller, they might approve my Apple Watch in *checks note* 2027.
This is getting better and better.
By now I have daily calls with the ATT rep who tries it's hardest to get my 3$ Apple Watch plan addition approved.
Fun fact he knows OpenClaw and uses it every day.
🙃🦞
I now have a working Apple Watch Cellular and a bank account.
Peter Steinberger reposted
If you missed the roundtable discussion with @zachlloydtweets, @steipete, @DynamicWebPaige, and @andrew__reed, you can get access to the recording here 👇
www.warp.dev/events/ai-engineer-after-hours-crafting-soft…
Never thought I give @Steve_Yegge a shoutout. He was just early, like most visionaries. Now everyone is building factories.
Peter Steinberger reposted
something something slash goal
The OpenAI booth is just straight up playing the match to get people to the booth and it worked lmao
Peter Steinberger reposted
I want to introduce myself.
Hi I’m Colin, I did a lot of the @openclaw iOS UI, so as one of the volunteers responsible, I accept my public flogging.
I’m just a maintainer with limited time and too much excitement about mobile agents.
Roasts fair. Have a dialog with me tell me it’s bad and how to make it better. Come join me in the discord so we can make it better as a community.
I’m looking for a semi-private hack space for a few days in SF for me and some of the OpenClaw maintainers. We need to cook. DM leads my way 🙏🦞
Peter Steinberger reposted
Holy sh*t after only 5 minutes…
Peter Steinberger reposted
In all seriousness Fable 5's guardrails seem much weaker now. This thing is straight up ripping through sandboxes.
I got Fable 5 to do dangerous cyber things. It found some major vulnerabilities in an agent sandbox.
Who in the US government do I call?
Pointed codex at some Twitter feedback on the OpenClaw iOS app and it did a first improvement pass. It's still not good, but for two prompts it aint bad.
Especially cool how it uses computer use to add before/after screenshots, as there's no GitHub API.
github.com/openclaw/openclaw/pull/98452
Peter Steinberger reposted
On price/token != cost/task:
To help internalize this fact for yourself, run Terminal-Bench using Haiku and then using Opus.
Here are the results for a 15-task subset. Haiku is 10x the cost!
Price per token != cost per task
Peter Steinberger reposted
Enjoy the codex reset people.
A user got to push the secret button and suddenly all of you got more tokens.
Hope the button goes back to @thsottiaux where it belongs.
(Also I think @sama has one)
Asked codex to download+transcribe all sessions from @aiDotEngineer and tailor them to my interests.
Speaker segmentation and transcripts are all done by summarize.sh
Peter Steinberger reposted
Episode 2 of OpenClaw’s official podcast
@steipete and @somalley108 join @hrudolph and @Pat_Erichsen to discuss OpenClaw stability, security, the future of OpenClaw, and the community’s response to the recently released OpenClaw mobile app.
www.youtube.com/watch?v=VJMhxh7KpqQ
Peter Steinberger reposted
There's a lesson here for everybody who thinks that "I'll just use a cheaper model for certain tasks"
It's *very* hard to know in advance how smart a model has to be to do a task.
If it's not smart enough, it will likely keep trying, using cheaper but more tokens.
Oh my god, Sonnet 5 was MORE EXPENSIVE THAN FABLE to run the whole bench 💀
Will be talking about the "app situation" and roadmap plans with @hrudolph live TODAY at 11:30 AM PST on the official OpenClaw podcast, live on Discord. discord.com/invite/clawd?event=1518848653291163728
sneaky, but also clever. thereallo.dev/blog/claude-code-prompt-steganography
🙃
Peter Steinberger reposted
"Fable 5 will be included for up to 50% of weekly usage limits through July 7, after which it will be available via usage credits" 🪦
Price per token != cost per task
Apparently we didn't talk enough about w̶o̶r̶k̶f̶l̶o̶w̶s̶ loops yet! See ya there!
@steipete is now joining us for Crafting Software Factories!
📅 6pm Wed evening in SF after the @aiDotEngineer World's Fair
He'll be joining @zachlloydtweets and Paige Bailey from Google DeepMind to chat all-things loops, software factories, and the future of coding
We moved the changelog to our docs to be easier readable.
v2026.6.11 has dropped.
This release focuses on the rough edges that make OpenClaw feel less dependable: misplaced replies, stuck sends, reconnects, model setup failures, and more.
Beware, this release is boring.
docs.openclaw.ai/releases/2026.6.11
Folks I found the reset button. @thsottiaux
Honored to be part of @aiDotEngineer’s keynote today!
As agents take on longer-running work, engineering shifts to setting direction, reviewing work, and designing better systems around the models.
@steipete at @aiDotEngineer
Peter Steinberger reposted
gpt 5.6 at 750 tok/s doing computer use is going to be a little scary
Peter Steinberger reposted
what’s a little funny about the “GPT weak on frontend” discourse is that everything we ship in the codex app gets adopted by the entire industry within days or weeks, pixel for pixel
If you want to add a navigation trail to your own chat app, the new MessageScroller component has the hooks you need out of the box. Look for:
const { currentAnchorId, visibleMessageIds } = useMessageScrollerVisibility()
Peter Steinberger reposted
I think the best thing about watching @steipete talk is that you can tell he LOVES what he does.
Peter Steinberger reposted
Day 2 at @aiDotEngineer World’s Fair. Great talk by @steipete on building loops.
Good morning @aiDotEngineer !
Are we talking loops today?
Was thinking if I should highlight this tweet or not, but it’s a masterclass in the amount of vitriol people face when working on open source.
Is the app great yet? No. It’s a start.
It was built by the community. Getting the iOS and Android apps working with secure pairing and push notifications - and getting both through App Review -took a surprising amount of work.
OpenClaw wasn’t acquired by OpenAI and isn’t an OpenAI product. It’s an open, independent project under the OpenClaw Foundation. OpenAI sponsors the project’s token usage; I work there.
Cristian, your tweet was just one of ~30 I woke up to today. I’d genuinely love your help making it great.
Attention is still the scarcest resource.
I’d rather spend mine encouraging people who build.
imagine getting acquired by @OpenAI, get unlimited AI tokens and still drop this slop abomination
Peter Steinberger reposted
My NEW Codex workflow is better than I expected.
8 new features ready for release in my app.
Took a few attempts to figure out the workflow and some bugs. Feels like the future.
Thanks @Dimillian @emanueledpt and @steipete for the inspiration.
Peter Steinberger reposted
Fun fact: The OpenClaw iOS and Android app (released today 👀) collects less data than most weather apps.
Nice!
was this vibed
This would have been amazing a few years ago, but in the age of AI, what’s the point?
Apple acquires Play. Design Mode in Xcode incoming.
Peter Steinberger reposted
Obviously yes.
As engineering, product, design, DS, etc. melt into a new kind of role, I was reflecting on what roles might look like in the future. For example, when I look at the Claude Code team I see what I think is five archetypes:
1. Prototyper: comes up with brand new ideas; churns out many ideas, most of which don't ship
2. Builder: quickly turns a prototype/idea into production-grade product/infra
3. Sweeper: cleans up the UI, simplifies the code and system, unships, optimizes performance
4. Grower: takes a product that has been built and iterates on it to improve Product-Market Fit
5. Maintainer: owns a mature system to make it secure, reliable, fast, and efficient as it scales
Many people span across 2 roles, and sometimes 3 roles. I also notice that these roles are not really tied to job function -- eg. across Anthropic, some designers match category 1, some 2, some 3; same for engineers, PM, DS.
A healthy team needs a mix of these, depending on the product:
- A product that is new and pre-PMF needs people that are strong at 1+2+3
- A product that is growing and has found PMF needs 2+3+4 and some 5
- A product that has strong PMF needs 3+4+5 and some 2
Maybe product roles of the future will look more like this, and less like the domain-specific roles of today?
Peter Steinberger reposted
OpenClaw is now on iOS + Android 🦞
📱 Native mobile apps, finally
💬 Agents in your pocket
🔔 Channels, tasks, replies on the go
Run agents from wherever your thumbs are.
iOS: apps.apple.com/us/app/openclaw-ai-that-does-things/id6780…
Android: play.google.com/store/apps/details?id=ai.openclaw.app
Peter Steinberger reposted
AI agents may end up doing more for web accessibility than accessibility laws. Agents rely on accessibility APIs to use computers and websites efficiently. As websites adapt for agents, people with disabilities benefit too.
Peter Steinberger reposted
Codex desktop is miles ahead of Claude for remote access via mobile or another computer.
Codex allows you to see **all** running sessions on **all** devices from any other device, including mobile
You can truly go touch grass with Codex. Brings me joy every day
Peter Steinberger reposted
friend's codex after talking to my codex
"History teaches us that access blockage rarely stops determined users." www.chinatalk.media/p/how-to-buy-cheap-claude-tokens-in
Peter Steinberger reposted
So interesting how many brilliant mathematicians insist that they are not especially intelligent or quick. Never sure what to make of this.
"I don't have any magical ability. I look at a problem, play with it, work out a strategy."
— Terence Tao
Got excited about the 52’ Dell, but it requires BetterDisplay hacking to get a usable resolution, but then I get weird flickers and also startup wake issues. [on M4 Max MBP]
So, Apple wins this round, 2xXDR it is.
www.dell.com/en-us/shop/dell-ultrasharp-52-thunderbolt-hu…
Peter Steinberger reposted
When @steipete launched @openclaw I was so amazed by all the interesting deep use cases for personal use.
For me though was “these deep use cases are cool but how can we do this wide instead?”
After trials and tribulations with ONE gateway and multiple users (please don’t 🤣). I ended up putting those Kubernetes and infosec skills at work and… well
I wrote this overview of how we’ve deployed it to over 500+ employees and it has been such an interesting ride with all the use cases that employees are coming up with
We getting Apple Car after all
NEW: Apple and Audi alumni just unveiled a $25,000 open-air electric neighborhood vehicle.
Amble One is a street-legal EV built for short local trips, with no doors, fewer screens, and a modular design inspired by the 1960s lunar rover.
• Goes up to 40 mph with 60+ miles of range
• Weighs under 1,000 lbs with a 5-hour home charge
• Rear seats fold flat for cargo, surfboards, or gear
• Built-in mounts let you add baskets, straps, mirrors, and cargo accessories
• Already has 500+ vehicles committed by resorts
Reservations are open now for 2028 deliveries.
wouldn’t that also make the tools better for humans
I really want CLI tools to start speaking directly to coding agents with errors. Instead of:
“Error:”
Return:
“Here’s what broke. Here’s how to investigate. Here’s how to generate a redacted repro. Here’s where to send it.”
Obviously there are privacy/security challenges to solve. But if we get this right, we can have a virtuous software improvement loop. Every failed agent interaction becomes a high quality bug report. Agents find the bugs. Agents fix the bugs. Software gets better faster.
Peter Steinberger reposted
I love how Apple notarization breaks multiple times a year until I manually log in and accept some new legal agreements.
Peter Steinberger reposted
What does the next training paradigm look like?
0:00:00 – The big research bet the labs are making
0:02:12 – Grindability is just as important as verifiability
0:06:10 – Will RLVR alone generalize?
0:08:41 – Getting the learning back to the weights
0:15:22 – Dreaming
0:17:23 – What 2027 looks like
Also on YouTube, pod feed, and Substack.
Peter Steinberger reposted
I stole this idea and now use it with every single employee.
It’s the best illustration I’ve seen of teaching someone to be high agency.
It says there are 5 levels of work:
Level 1: “There is a problem.”
Level 2: “There is a problem, and I’ve found some causes.”
Level 3: “Here’s the problem, here are some possible causes, and here are some possible solutions.”
Level 4: “Here’s the problem, here’s what I think caused it, here are some possible solutions, and here’s the one I think we should pick.”
Level 5: “I identified a problem, figured out what caused it, researched how to fix it, and I fixed it. Just wanted to keep you in the loop.”
Using this framework, here’s what I say to every new employee…
You will live at Level 4 from Day 1 and as we build trust you will rise to Level 5.
Being high agency doesn’t just mean tackling problems in this way. It means your entire way of working should be oriented to being a Level 4+ employee.
Plz feel free to steal it as well.
And ty @stephsmithio for the framework!
Peter Steinberger reposted
Post your best codex billboard
Peter Steinberger reposted
the one where @benthompson discovers codex stratechery.com/2026/my-vibe-coding-adventure-the-app-and…
Peter Steinberger reposted
"Dario Amodei hit a safety filter, and the conversation was automatically switched to Tom Brown"
truth
“Six months ago, my bottleneck was tokens.
Then I joined OpenAI.
Now my bottleneck is attention.” - @steipete
Peter Steinberger reposted
OpenClaw or BrokenClaw?
First off, I’d like to introduce myself. My name is Hannes Rudolph, and I’m the new Community Manager here at OpenClaw. Nice to meet you.
I want to acknowledge the frustration many of you have felt with OpenClaw’s stability. We’ve broken your claws, and that sucks. We’ve made some real progress recently, but there’s still work to do and we’re going to keep at it. We understand how frustrating it is to deal with repeated issues, especially when our public response has been lacklustre and hasn’t matched the seriousness of the problems you’re experiencing.
For anyone who wants to look at one data point showing how stability has changed over time, there’s an independent public tracker here: isitstable.iclaw.digital. We didn’t make it, and we don’t think any single tracker is infallible or tells the whole story. But it does help show how bad things got and where we are now. It’s a useful reference point, not a substitute for listening to people directly or looking at the issues they’re still having.
OpenClaw is nothing without the people who use it, build with it, question it, and help shape where it goes. We know that because many of us were frustrated users before joining the project. We care about this community, and we want this community to be a place where people can get help, share ideas, and be honest about what is and isn’t working. That means we need to listen when people are angry, engage when things are confusing, and make it easier for people to participate rather than shutting them down.
Please keep being vocal about what isn’t working. We’ll do our best to route feedback to the development team quickly and with the importance it deserves. I can merge things when needed, and we’re not interested in hiding behind process while users are dealing with broken or frustrating behaviour.
We’ve heard you. Now we need to show that we have.
Hannes
Peter Steinberger reposted
SoftBank’s investor presentation is one of the greatest things ever made. I’ve been thinking about it all day. These are the real slides shown in a speech where Masayoshi Son said he wouldn’t retire for at least another decade. The goose stuff is perfect.
group.softbank/media/Project/sbg/sbg/pdf/ir/investors/sha…
Peter Steinberger reposted
Check out episode 1 of The Clawcast, our official OpenClaw podcast, with @hrudolph, @Pat_Erichsen, and @GosuCoder!
Great convo around skills, Clawhub, securing OpenClaw deployments, and more.
youtu.be/IfJJnR1LIE0
Peter Steinberger reposted
🦞 OpenClaw 2026.6.10 just dropped.
Just a small release to keep things brewing:
⚡ Automatic fast mode for short talks
🧠 Much more reliable model routing
🔒 Safer session state + trusted policies
🛠️ Better provider onboarding
Helping deliver rock-solid lobsters. 🦞
Peter Steinberger reposted
Well @cognition were right about Devin. And @steipete was right about Personal Assistant agents. And @GeoffreyHuntley was a year ahead on loops. The future always looks a bit messy when it first arrives.
Google fired the guy that made the google workspace cli, because he made the google workspace cli.
Lucky me, Google can't fire me. gogcli.sh
Two months ago I was fired by Google for creating the Google Workspace CLI. It went viral, hit #1 on Hacker News, gained thousands of GitHub stars and many thousands of actual users in just a couple days.
It was an incredible, confusing journey, from directors and leaders asking what they could learn from the tool to getting grilled by legal about why the Google logo and brand colors are on the Google Workspace GitHub code repositories.
I think the cause was that Workspace and certain leaders (and projects) were afraid of being disrupted. But the fear wasn't specific to my CLI, it was a broader fear in what agents meant for Workspace. Either way, the irony of my termination was the announcement at Google Cloud Next two days before I was fired that an official Workspace CLI was coming.
I want this out there because it is easier for me to explain my story and it is an experience I want to fully own. It's also part of my healing.
Nearly 7 years at Google was an incredible opportunity for me and I was fortunate to have wonderful teammates and a manager that fully supported me through these last few months. Thank you.
I feel the same about shared office spaces.
Why Remote Work is White Collar Fraud.
"I have a three-year-old and a five-year-old. The idea that I could do any work at my house is like a total fantasy.
The kids come home at 3pm, your work day needs to keep going. I'm highly against it." @typesfast
Peter Steinberger reposted
another Codex for OSS update:
Oh My Zsh
Ollama
MarkItDown
Langflow
scrcpy
LangChain
llama.cpp
shadcn/ui
RustDesk
three.js
v2rayN
Browser Use
Storybook
Home Assistant
Svelte
fzf
Hoppscotch
Netdata
Dear ImGui
LlamaFactory
Tabby
DeerFlow
Anime.js
Ansible
ripgrep
openpilot
Spring Framework
Ghostty
whisper.cpp
Expo
Julia
tmux
Apache Airflow
Milvus
Gradio
DaisyUI
Dioxus
Hyprland
Drizzle ORM
FreeCAD
Hono
thank you for building the software we all use
Peter Steinberger reposted
/goal make postgres multithreaded
....
1k commits
124k lines
786 changed files
10 days later
Patch the Planet.
Patch the Planet is our effort to help open source maintainers move from security findings to merged fixes.
We’re working with Trail of Bits, HackerOne, Calif, researchers, and maintainers to bring Codex Security and advanced models into the remediation process, with human review at the center.
Peter Steinberger reposted
Why do people think there is more clarity in their communications when they throw their thoughts into an LLM before sharing? I’m so sick and tired of engaging with people who send me regurgitated clanker slop :-/
This is becoming my favorite way to read Twitter. birdclaw.sh
Peter Steinberger reposted
to be clear, this is a closed source orchestrator on top of closed source models. if before you didn't control the models, now you don't even control which ones are used or how much. this is not "AI sovereignty"
i've also read the tech report to get an opinion on the technical stuff:
fugu (not the ultra version) is basically a classifier that selects which model at each turn is most likely to answer correctly (in other words a router). this leads to -10 points on SWE Bench pro compared to opus, gets some gains on other benchmarks but very slight. argument could be that it reduces cost, but no information about this so it's likely the opposite. they also have an autoresearch benchmark where they compare to frontier models "Model A, B and C" which is really crazy to not be transparent about what models you compare against. let's also say that this probably doesn't support adding new llm out of the box since you need to retrain the classifier
about fugu ultra, this is basically and advanced plan mode and orchestrator, this is a model that for a query outputs a plan with multiple "workflows". my understanding of workflows is that they say: "spawn model A subagents to achieve this, then use model B to judge it, then summarize this with model C" which is just a test time scaling compute strategy. i think this is an okish way to do it, but it's limited by the fact that they need to predict everything before the agents start working, which is why they limit this to 5 steps. imo you need to predict what to spawn at t+1 with the information you get at t, not with the info you get at t=0. there are also other issues such as fable 5 score on terminal bench being wrong and them being super vague and unclear about which model is in the LLM pool (they only mention closed source api one)
the biggest and most obvious issue is that they are introducing a "test time scaling" method with "best of N" over models, and they literally NEVER REPORT the number of output tokens or cost to achieve a benchmark/task
the good comparison here is not with opus, but it's opus with ultracode/workflows enable, not with kimi, but with kimi swarm ect.. very very confusing release
Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API.
Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls.
Try it: sakana.ai/fugu 🐡
People here discussing what happened with OpenClaw.
The hype died down. We improved quality and grew a team. We created a non-profit whereas competiors are VC funded and have other agendas.
This is our strongest week so far.
New NPM all time high record for @openclaw, going from strength to strength!
I was skeptical about the multi-model routing. Seems my hinch was right.
I tried this so you don't have to.
I know this is going to absolutely shock you but no this does not match the performance of Mythos.
A few early thoughts:
1. The limits are pretty bad. I used 100% of my 5-hour usage in less than 1 prompt.
2. I specifically gave it a threejs task because it is an area that SOTA models have made big strides in, that other models just are not great at.
I asked it to build a replica of Rocket League. I'll put the prompt in the comments.
The game was pretty bad and notably worse than GPT 5.5.
Even after multiple fixes, it took 7-8 back and forth with Codex just to get it an almost playable condition. Prior to these fixes, the game was not playable.
Maybe it's really strong in other disciplines. I'd love to test that but I hit my limit in 1 prompt lol.
GPT 5.5 by contrast did a pretty good job and required no follow ups. Fable would have absolutely nailed this as well.
But yeah, early impressions...not great. But I hope I'm wrong. More testing tomorrow.
Introducing Sakana Fugu: A full multi-agent orchestration system accessible via a single model API.
Our ‘Fugu Ultra’ model matches the performance of Fable and Mythos, delivering frontier capability without the risk of export controls.
Try it: sakana.ai/fugu 🐡
Peter Steinberger reposted
Some thoughts on ownership I shared with the team this morning
Peter Steinberger reposted
building a company is largely the act of managing your own emotional state so the field around you remains coherent enough for other ppl to do their best work inside it.
basically founders set the emotional physics of the org. panic compounds. but clarity compounds too.
Peter Steinberger reposted
Another day and more tooling, Clawrouter under MIT license is a @cloudflare based rust WASM that allows ephemeral single API key for agents and team to access a huge range of model providers with a single auth or GitHub login. github.com/openclaw/clawrouter
Make that side button useful:
Peter Steinberger reposted
I used to be a die-hard Claude Code user.
Codex has won me over because:
→ GPT-5.5 is excellent
→ Fast mode + generous limits = more reps
→ Little touches like steering, auto remote control on phone, etc
But most of all Codex's browser and computer use capabilities are simply goated. I built so many workflows relying on those two things alone instead of hunting for APIs.
I still use Claude Code too. The app seems to be getting better and the design and frontend capability of Opus is still much better than GPT. Whenever Fable comes back that's another reason to go back.
Honestly, I hope these two compete forever and other players (Cursor/Grok, Gemini, etc) all stay competitive.
This way the builder keeps winning 🙂
Peter Steinberger reposted
There’s an OSS implementation of the codex permission flow in Swift here:
github.com/zats/permiso
great design in 2026 = make easy what the platform tries to block or make hard
to enable computer use, Codex opens macOS settings, overlays this right underneath the list of allowed apps, and lets you drag & drop
reminds me of when people figured out things like the double-confirm opt-in pattern for push notifications on iOS last decade
If you are in Japan🇯🇵 or do business there, here's your chance to get lots of tokens!
Startup founders - @OpenAI and @OpenNetwork_Lab are hosting a Startup pitch contest in Kyoto with OpenAI providing $1M USD in API credits across the program, including $100,000 USD for the 1st place winner! 🇯🇵
Apply NOW ! 👉 luma.com/wdfs5azz
Hannes speaks both developer and agents. Blessed to have him on the team!
Peter Steinberger reposted
Codex can now hand off threads between local and remote hosts.
Start work on your laptop, send it to a remote box before you close the lid, bring it back later.
And yes, Codex can orchestrate the handoff for you.
Peter Steinberger reposted
If you're on your way to building a billion dollar company that involves a web app, here are some of my notes on architecting the frontend.
if you don't do this, it's probably fine but one day you'll hire someone to fix it but truly that person could be doing some other higher value thing if you make some key optimizations on day 1
you don't even have to learn anything you're gonna tell your agents to do it anyways!
okay here it goes:
- Make your server code generate a openapi spec which then generates all the relevant client side code. Never do this by hand. Typing backend types instead of generating them should be banned
- You need to make a decision on how the client talks to the backend. rest/graphql works in which case please just use tanstack query. other libraries will look similar but tanstack query truly is goated.
- if you want linear style sync setups or offline mode, think about this HARD and architect it from day 1. Bolting this on later is so tedious.
- People like using plain react router but things have gotten a lot better since then. Try their new framework mode or just even use tanstack router. Use route data loaders.
- If you store a lot of state in query params, make that a first class citizen and make sure its type safe. use nuqs or tanstack query.
- Most apps just need a single state management situation for server state and thats it. If you have other bespoke needs, i have quite like zustand and xstate/store.
- If you have a super interactive app where things come in and out of view, theres a lot of frontend state to maintain, music is playing and what not, lock in and learn xstate. Trust me if you wanna keep ur sanity, you need to model ur frontend as a state machine otherwise you're gonna be deep in useEffect hell
- React compiler is here my friends, the days of useMemo and useCallback are gone. Update your priors accordingly
- Tailwind is easy and fun but makes it really hard to maintain a large app with consistent styling. You need a "agent-first design system/component library" but maybe this is a rant for another day
- Don't be afraid to hack your routing library to fit your needs more closely. A lot of apps have "drawers" to show additional info. You should 100% be able to say "here's a route, make it a drawer" and everything should be handled from there.
- Managing loading and error states using isPending and isError is madness. Lean into Suspense and ErrorBoundary.
- Figuring out a blessed path for websockets and SSE on day 1 i think will pay dividends in the long term if you're building anything AI related.
- If you're building a SPA, don't use next.js. it literally makes no sense. Why would you do this.
- Definitely deploy on Cloudflare or vercel. There are other services but trust, there have weird missing features.
- Assuming you build something people want, the next job is to build the factory so it can efficiently build the thing. Act accordingly.
Peter Steinberger reposted
If you're curious how I managed to do over $20,000 in inference on the last 48 hours, here's a video all about it.
Spoiler: loops are really powerful
Everything’s either a fast or slow API now.
The best way to think of Record & Replay™️ is to imagine it as a new primitive
Ask Codex to watch you do something on your computer -- anything in your day-to-day
And Codex will process the (1) recording and (2) events (same spine as computer use) to understand (A) exactly what happened, and then (B) turn it into a skill you can use for the future
Some ways I've been using R&R:
- formatting calendar invites in my own particular fashion (I had Codex watch once, and now it matches my pattern when using the gCal plugin)
- drafting PRs and sharing them to the right slack channel with the exact formatting I like
- testing onboarding flows (Codex can pickup on different friction points during the recording -- for this, I don't need it to create a skill)
Peter Steinberger reposted
I love Codex automations. A useful trick is to give them two loops: an inner loop that does the work, and an outer loop that uses your review to improve the next run.
That way, the context you add today isn’t lost tomorrow.
More here 👇
Peter Steinberger reposted
The infamous "The compiler is unable to type-check this expression in reasonable time" error should be a thing of the past.
Apple has reworked how views are built under-the-hood so that type-checking becomes easier for the compiler.
Peter Steinberger reposted
Amazing things are happening with Codex Mobile, build & run apps directly from your phone!
Introducing Sqim - a free tool that lets you sign and upload your iOS project binaries, and serve temporary web pages to sideload on your iPhone straight from Codex Mobile. This closes the feedback loop in your mobile agentic iOS development workflows on any network without requiring tailscale, VPN, or simulator streaming hacks!
Peter Steinberger reposted
Meta HQ right now.
JUST IN: Meta’s CTO says morale is near “the worst it’s ever been” — leadership will offer increased snack budgets to lift spirits.
Peter Steinberger reposted
Creator of Sqlite on pull requests: "You say, oh, it's free. No. It's not free. What you're doing is asking me ... to maintain it for you, to to document it for you, to test it for you, to maintain it for you for the next 25 years. That's not free." Yep.
Wise words from a wiser man than me. I've told people for the past decade and I have recent posts on here saying the same: the merge button is the easy part. Its the decade+ (Richard says 25 years) that follows where you've accepted the transfer of maintenance thats hard.
Peter Steinberger reposted
GLM 5.2 is absolutely convinced that it is actually Claude, from Anthropic. When I tell it that it's GLM 5.2, it refuses to believe me, but is willing to check the local agent config to see what model is running.
The realization:
Peter Steinberger reposted
At OpenAI, we're continuing to bet on Rust as the future of systems programming.
I'm proud to announce that we're making a $600,000 commitment to the Rust Foundation, which combines our Platinum membership with additional support for maintainer efforts across the Rust ecosystem.
Peter Steinberger reposted
PR limits are live 🚀
Repo admins can now cap how many open PRs a user without write access can have at once, with a bypass list for trusted contributors.
A big step toward helping maintainers manage contribution volume and reduce PR noise.
Peter Steinberger reposted
new shape-rotator benchmark
Fable and GPT-5.5 of course far ahead of the field
but now look at GLM-5.2. it's ahead of Gemini 3.5 Flash and Opus 4.8
you can't really benchmaxx a benchmark that was just released
so the GLM-5.2 gains seem more and more like a genuine improvement!
Are AI agents shape rotators? In this new benchmark, we let the models play campaign puzzles in Opus Magnum, a puzzle game by @zachtronics.
Ironically, Claude Opus 4.8 performed poorly, being beaten by GPT-5.5, Gemini 3.5 Flash, and GLM 5.2. Claude Fable 5 crushed them all.
Peter Steinberger reposted
Gigalong $SNAP. Absolute game changer for boring meetings!
Peter Steinberger reposted
Reminder that you can use the Codex App, CLI and SDK with any open source model, not just with OpenAI models.
developers.openai.com/codex/config-advanced
Peter Steinberger reposted
I was walking the same path:
• “I need remote CLI for the gym”
• “Actually I need an agent for this project”
• “Actually I want an agent managing all my projects”
• “Oh and also it must learn our workflows as they evolve”
The moment I gave my agents all my creds I realized how you came up with OpenClaw
Peter Steinberger reposted
babe wake up, new Fullbody Ultrasonic Computational Tomography device just dropped
sci-fi vibes intensify
Announcing a new division of Midjourney called "Midjourney Medical"
Peter Steinberger reposted
🦞 OpenClaw 2026.6.8 just dropped.
💬 Richer Telegram + WhatsApp
⚡️ Sharper agent & Gateway recovery
🧠 New models + tougher memory
📊 Native /usage footers
📱 Smoother WebChat & iOS
Let’s dig in!
github.com/openclaw/openclaw/releases/tag/v2026.6.8
Peter Steinberger reposted
Codex now supports .worktreeinclude! Worktree enjoyers can now automatically copy over things like .env files, config files, agent overrides, etc.
developers.openai.com/codex/app/worktrees
Peter Steinberger reposted
Two years ago, reasoning models did not exist. GPT-4o and Sonnet 3.5 were SOTA.
One year ago, the best publicly available model was o3, and 99% of those who used it (i.e., 99% of 7% of OpenAI customers) were using it as a chatbot.
Today (i.e., right now as I type this), I have Codex building several pieces of personalized software and doing a large-scale data management project for me. The data management project has been running autonomously for nearly 4 days. I have multiple working apps on my phone that I use daily and which were built entirely by Codex.
Projecting this trend to June 2027 is absolutely mind-boggling. The world will change.
Peter Steinberger reposted
just two guys talking about loops
Peter Steinberger reposted
Some observations from disassembly that might be related to iOS 27 performance boost
codecolor.ist/posts/2026-06-15-ios27-reworked-stub-islands/
Peter Steinberger reposted
NEW: Inside Cursor's wild rise. Lots of great new details:
• CEO Michael Truell didn't pay himself for years
• Cursor once made up 40-50% of Anthropic's revenue
• Anthropic told Cursor that Claude Code was just a 'research effort' (lol)
• Cursor's unpaid "work trials"
Peter Steinberger reposted
Open Source runs on maintainers!
at @OpenAI, we're committing $160,000 to sponsor the maintainers behind the Astral and Codex toolchains
This is alongside our ongoing $1M fund providing free Codex access to open source maintainers.
I’m incredibly stoked that this finally happened and as someone who owes his career to open source I’m excited to continue doubling down on this through the year!
Huge thank you @thsottiaux, @charliermarsh, @romainhuet and the codex team
Lots more to come on this front, stay tuned!
Proud to announce that @OpenAI is renewing and expanding support for maintainers of open-source projects used across the Astral and Codex toolchains.
Today, that starts with committing more than $160,000 in direct sponsorships through @GitHub Sponsors.
Peter Steinberger reposted
Anthropic's Safety Superpower
Anthropic's belief in its own commitment to safety gives the company license to aggressively favor its business and even challenge the U.S. government.
stratechery.com/2026/anthropics-safety-superpower/
Wanted to buy a new Mac Studio for San Francisco.
We don’t even know how to make these anymore.
Whenever you create an issue on one of oure open source projects, @clawsweeper will review it, and *if* it fits the VISION.md file, will pick it up and create+autoreview a PR.
e.g.: github.com/openclaw/gogcli/pull/816
IS THIS A LOOP
Peter Steinberger reposted
Codex hit Figma's MCP limit and then just opened Figma in a browser tab to keep working
Mosh + tmux/zellij is a lifesaver for bad in-flight internet. (i prefer zellij but either works)
Peter Steinberger reposted
I'm doing one quick targeted pass so I can give you the exact answer rather than hand-waving
Peter Steinberger reposted
The Rio 3.5 model broke the internet this week. The plot twist? It’s essentially our open-source model, Nex N2 Pro, wearing a different hat.
🤯 We analyzed the weights, and the recipe is exact: Rio 3.5 ≈ 0.6 * Nex N2 Pro + 0.4 * Qwen 3.5
It even literally introduces itself as "Nex N2 Pro" if you ask it without initial system prompt!
😂 We are flattered that the City of Rio used our work to achieve SOTA performance. Thanks for the ultimate benchmark validation.
🤝 But in the open-source world, attribution matters.
👇 Full mathematical proof & verify script in the first reply!
Peter Steinberger reposted
I basically never write my own /goal anymore.
I ask Codex to write one for itself, and one for each agent it spawns.
Like this 👇
This shortage of chips is getting out of hand.
Today on the blog, we discuss a pathway for the second life of phones through the exploration of “phone cluster computing”, which can directly reduce the environmental footprint of computing by avoiding the need for further raw material extraction. More →goo.gle/4aJe5vO
Peter Steinberger reposted
Add this shortcut to your iOS Home Screen so you can search for Codex. And use this icon raw.githubusercontent.com/lobehub/lobe-icons/refs/heads/m…
www.icloud.com/shortcuts/857005fe2b7d47e5bcd8f2147050bfcb
Got a PayPal verification text and thought I been hacked, but it was just codex signing up for a web service it needed.
Peter Steinberger reposted
I can barely keep up with implementing/testing/landing all the Issues/PRs folks submit to github.com/openclaw/crabbox
Codex runs INSIDE crabbox while it is building crabbox.
This is becoming essential infra for my work.
Codex been looping nonstop for the last 4 days in multiple trees. Since all of it is e2e verifiable it basically builds itself. Codex even signs up for the services automatically via browser/computer use. My main job is adding credit card details and closing things that I don't see as a fit.
Peter Steinberger reposted
Dario (48 hours ago): “US gov should be able to block model deployment”
USG: *export controls models*
Dario: “not like that”
The Trump administration has placed Anthropic's Mythos 5 and Fable 5 under export controls. Commerce Secretary Howard Lutnick sent a letter to Dario Amodei tonight stating that foreign governments, companies, and individuals will no longer have access to either model.
Peter Steinberger reposted
新しいリアルタイム翻訳モデルを発表できることをうれしく思います。ぜひ本日よりAPIでお試しください。
“not consistently candid in their communications” is my fav new americanism. www.theinformation.com/articles/anthropic-blindsides-busi…
Peter Steinberger reposted
appshots in codex is the most useful piece of software on my Mac
most of my prompts these days are:
- cmd + cmd investigate this
- cmd + cmd open a PR to fix this
- cmd + cmd run the eval on these set of prompts and discussion
- cmd + cmd set a heartbeat to keep following up with this
and many more cases
trust me, you’ve got to try it
codex -C ~/projects/openclaw -m gpt-5.5-cyber time
Peter Steinberger reposted
more like Fumble 5
How am I only now finding out about appshots?
I was dragging screenshots into codex live a caveman.
Peter Steinberger reposted
We've updated the Artificial Analysis Coding Agent Index, replacing SWE-Bench Pro with Datacurve's DeepSWE benchmark - the swap lifts Codex with GPT-5.5 (xhigh) above Claude Code with Opus 4.8 (max), while the newly released Claude Fable 5 (max) in Claude Code debuts at the top
DeepSWE, built by @datacurve, writes its tasks from scratch rather than adapting them from public GitHub issues or pull requests, so no model has seen the solutions during training. That matters because SWE-Bench Pro, the benchmark it replaces in our Coding Agent Index, had grown gameable, with some models recovering the fix from the repository's commit history instead of solving the task.
The swap reorders the index: Codex with GPT-5.5 (xhigh) rises from 65 to 76, overtaking Claude Code with Opus 4.8 (max) at 73. Claude Code with Fable 5 (max), which enters directly on the refreshed index, leads at 77. SWE-Bench Pro had been flattering some combinations and penalizing others.
More below.
Peter Steinberger reposted
Everyone says the latest AI agents will be "job-ready" soon, especially after the release of Fable 5 this week. But is that really the case?
Over the past many months, my group and collaborators have been building Agents' Last Exam (ALE), a benchmark designed to test exactly that claim on real digital labor-market work.
My group and collaborators previously have created many of the benchmarks the field runs on, including MMLU, MATH, CyberGym, and ExploitGym. Today, I'm excited to share Agents' Last Exam (ALE): a rolling benchmark that measures whether AI agents can actually perform economically valuable work across a broad range of real-world domains.
With ALE, we evaluated Fable 5, GPT-5.5, Composer 2.5, and other frontier agent systems across more than 1,500 expert-sourced tasks spanning 55 occupations.
The result is both impressive and sobering.
Today's agents can solve a meaningful fraction of professional tasks. But when we look at the hardest tasks, the ones requiring sustained reasoning, deep domain expertise, and reliable execution over long horizons, they are still far from human-level performance.
On ALE's hardest tier, every frontier agent we tested, including Fable 5, achieved a 0% success rate.
The age of useful agents is here.
The age of truly job-ready agents is not.
We hope Agents' Last Exam (ALE) will serve as a new guidepost and north star for developing agents capable of reliably performing economically valuable work across a broad range of domains.
🧵
Peter Steinberger reposted
Fable 5 refused 200 out of 200 ProgramBench tasks lmao
IMO sth that is a bit overlooked but will become far more important in the future. GPT is 10-20x more token+cost effective for ~similar outcome.
Day 3 with Fable.
Gave a huge prompt to implement a feature across CLI, web server, and another server to both Fable and deep^2 in Amp.
deep^2 was done before I went to the gym. It stopped short. Sent another prompt. $20.
Fable ran for 1hr40min and cost $350.
Results:
They both understood the assignment and built the same thing. Maybe that's due to my prompt.
Fable's worked on first try. Well done.
Deep's looks correct but didn't work on first try.
$20 vs. $350.
I'm sure I could get deep^2 to make it work and we'd end up at, what, $40? While Fable is now at $457 after I asked some follow-up questions.
Peter Steinberger reposted
Ship of thesus
Rewrite the entire codebase for each new model release
Peter Steinberger reposted
My new favorite prompt: take a step back. Wonderful results
Peter Steinberger reposted
Damn how did this leak???
OpenAI is preparing a new AI model, per The Information
Peter Steinberger reposted
1. This is why @thsottiaux and @OpenAI are feeling good about GPT-5.6 (screenshot is from @datacurve's latest DeepSWE from @theo's video)
2. Most other SWE benchmarks are out the window. Anthropic admitted claude cheat(s) on swe-bench pro. I think DeepSWE might be one of the 5 most important AI benchmarks out right now.
3. GPT-5.5 was released a month before Opus 4.8, and it is still outpunching Anthropic's newest Fable 5.
4. In other words, Anthropic, the company known for their coding models, have had TWO additional shots at creating a coding model that competes with GPT-5.5 on a cost-per-intelligence basis and have failed.
5. A little secret most benchmark watchers overlook? GPT-5.5 xhigh is not even close to being OpenAI's best model. That belongs to GPT-5.5 Pro, which rarely gets benchmarked.
6. This will be very bad for Anthropic when/if Tech Twitter, the mainstream media and Wall Street start connecting the dots.
7. Seems like all the IPO doomsday marketing hype from Anthropic couldn't actually make their models better or more efficient.
8. The above is why Anthropic is now going on a PR tour with national news companies, as they see the writing on the wall: OpenAI's models are better at what matters and its Codex harness is better.
9. Anthropic's best shot at out IPO-ing OpenAI is only:
A) controlling the public narrative via press tours
B) confusing enterprise customers (agentic models running broken loops, tokenmaxxing and companies are none the wiser)
C) Copying Codex verbatim like Google tried to with Antigravity, b/c Anthropic's Caude desktop offering is a disjointed, siloed mess without unified memory.
Feeling pretty good about things
Peter Steinberger reposted
Our Anthropic overlords deciding which prompts the peasants are allowed to use.
Peter Steinberger reposted
2026
Peter Steinberger reposted
we shipped the Tibo button
Getting Chris to do a PR with Codex!
Cooking. @steipete 🦞
writing mac apps is still hard.
the WindowServer forensics (not mine, excellent work): github.com/steipete/CodexBar/issues/1399
my macOS 27 datapoint: github.com/steipete/CodexBar/issues/1427
the PRs: #1277 #1286 #1297 #1314
Peter Steinberger reposted
By the way, public service announcement: if you're one of the numerous people posting about Anthropic's dystopian ways and you're thinking about getting Claude to help you write that post... don't!
Another one of their terms is that you may not use Claude to do anything that "exposes [Anthropic to] reputational harms" 👇
And, if you do, under the - extremely unusual - clause 13 of their terms (www.anthropic.com/legal/consumer-terms), you have PRE-AGREED, by using Anthropic (and accepted their terms), that the harm you've done is irreparable, that you won't oppose Anthropic injunction, and they don't need to prove actual damage.
They can simply go to a judge in a friendly jurisdiction (and of course, their terms precise that any dispute "will be resolved exclusively in the state or federal courts located in San Francisco, California") and:
a) file an injunction that shuts you down
b) make you pay for everything since under section 11 of their terms you agree to indemnify Anthropic for "any and all liabilities, claims, damages, expenses (including reasonable attorneys' fees and costs), and other losses arising out of or related to your breach or alleged breach of these Terms."
In other words, if you use Claude to help you talk shit about Anthropic publicly, their terms say you pay their lawyers to go after you and you've already pre-agreed you've lost the case.
Oh, and cherry on the cake: in the odd case the judge were like "are you crazy, this is insanely abusive, you Anthropic are the ones at fault here," according to their terms Anthropic's maximum liability is... $100.
Peter Steinberger reposted
If you really think about it, despite being mocked as “ClosedAI,” OpenAI has contributed enormously to the field: GPT, GPT-2, GPT-3, CLIP, the ChatGPT paper, the GPT-4 Technical Report, the Sora technical blog, and even open-sourced Codex.
Anthropic, meanwhile, has contributed far less to the public research ecosystem while increasingly promoting fear-based narratives and restricting access through heavy gatekeeping.
The world I least want to live in is one where the future of AI is controlled by companies that prioritize secrecy, gated access, and centralized control over openness, reproducibility, and scientific progress.
Peter Steinberger reposted
This shows yet again how this limitation was never about "safety" but about Anthropic doing stuff just because they thought they can.
I am increasingly sceptical Anthropic really cares about safety, and not just their business interests (limiting competition where they can)
NEW: Anthropic is walking back Claude Fable 5's policy to covertly degrade performance for competing AI researchers, after facing fierce backlash.
“We’re changing Fable 5’s safeguards for frontier LLM development to make them visible,” Anthropic tells WIRED. “We made the wrong tradeoff and we apologize for not getting the balance right.”
Part of the OpenClaw hardening work is reducing surface risk; for some media conversion we had to shell out to ffmpeg.
In the next release that can now be done via wasm, with similar performance for our use cases.
github.com/openclaw/ffmpeg-wasm
Here's a simple loop: Tell codex to maintain your repos, wake up every 5 minutes and direct work to threads. That makes it easy to parallelize+steer work as needed.
I use a orchestrator skill combined with my triage+autoreview+computer use skills, so some work can land autonomously. github.com/steipete/agent-scripts/blob/main/skills/mainta…
github.com/steipete/agent-scripts/blob/main/skills/github…
Peter Steinberger reposted
Recently, we purchased one of each Anthropic/OpenAI subscription plan and randomly ran long horizon coding tasks until we exhausted the weekly limit. It's widely believed that a $200/month plan maxes out at ~$2000/month worth of tokens (assuming API pricing). However, we found that the subscriptions are actually far more generous. (2/4)
Peter Steinberger reposted
My last observation re: Anthropic’s secret sabotage safety policy, is that it undermines actually good safety policy. How?
1. First, it is very plausible to describe this as anti-competitive behavior (even if you are maximally sympathetic to Anthropic here you must admit this), and it is behavior being justified in the name of AI safety. If you believe, as I and many Anthropic staff do, that it may end up being critically important to relax antitrust enforcement so that the frontier labs can cooperate and collaborate on some areas of AI safety, Anthropic just undermined the case for that in a large way.
2. Overall, this massively and profoundly raises the status of the argument that AI safety has been hype to justify monopolistic behavior by labs. I continue to believe AI safety is a real and serious issue that is growing in importance rather than diminishing. If you agree with me, this incident is a setback, maybe a serious one.
3. As I have observed elsewhere, Anthropic’s official corporate policy is structurally identical to the fact pattern alleged against them by the Department of War. I still think DoW acted both falsely and wrongly in that fight, but it is no longer possible to defend Anthropic with a full throat after this incident.
4. This raises the case for heavier handed regulations. Anthropic is making an awfully good case here that their products ought to be treated as utilities, and thus that their alignment practices should be a matter of public policy rather than private property. I am starkly opposed to this sort of state power grab, but Anthropic is doing more to justify it than anyone else.
5. Thus, significant damage has been done to a community and entire approach to AI governance. It was done unilaterally by Anthropic, likely motivated largely by self-interest and justified within the internal psychology of the firm through the lens of safety.
I suspect this is fixable in the economic and legal senses for Anthropic, but I fear the trust that has just been broken, and the goodwill extinguished, will take very much time to repair.
Peter Steinberger reposted
Some thoughts of mine that nobody asked for on the EU's fight with Apple and my worries related to Mythos and Fable. lucumr.pocoo.org/2026/6/10/gaslighting/
Peter Steinberger reposted
iOS 27 framework references “foldState” and “angleDegrees” but I’m sure that’s nothing
Peter Steinberger reposted
🦞 Big security update for users and enterprise: Custom install policy for plugins and skills! 🔒
OpenClaw will call an executable of your choice with an extensive profile and source code of every plugin/skill.
You make the decision to allow install.
security.installPolicy lets operators approve or block skill/plugin installs with a trusted local command. It runs after staging and fails closed when enabled but unavailable.
docs.openclaw.ai/tools/skills-config
Peter Steinberger reposted
twitter feeds are looping these past few days
the pushback seems to be that it’s a workflow for the 1%
while there’s some truth to the claim, i think it’s applicable for everyone to different degrees
you can improve quality of any loop by spawning additional while loops at any part of sdlc
loop in moderation depending on where you are - token maxers can afford to loop endlessly on all stages in sdlc. but you can get 80% of the value by running targeted loops for a few rounds - eg. looping on a spec until major gaps are addressed and looping on code until there are no more critical review finding
Peter Steinberger reposted
Folks it is just control loop
Think of Kubernetes.
You write desired state/spec, what to observe/status, the controller, and the kubelet.
You let this whole control loop sense and actuate to prompt/drive the agent.
Obviously the right idea as AI gets smarter
Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.
You should be designing loops that prompt your agents.
Peter Steinberger reposted
can’t wait to see the new Ciri today
if there’s anything we can all agree on, it’s that the letter C is having a generational run
Peter Steinberger reposted
How do you use coding agents right now?
Peter Steinberger reposted
This is an intentional decision in our products. Same thing goes for ChatGPT for Excel, Sheets and Powerpoint. If you run out of limits in the middle of a task, why would we want you to have a half built spreadsheet or deck?
is it just me, or is Codex not stopping your actual task even though you're out of limits?
i've been on 0% for the past 5-10 minutes, and Codex continues to go work
surely this is a bug?
Peter Steinberger reposted
I have a new kind of big button that I can press for Codex. Over the next 100 days, we will select one person per day who does impressive or incredibly useful work with Codex and give them 10X usage limits for a month to see what they can do with it.
First one tomorrow.
Here’s your monthly reminder that you shouldn’t be prompting coding agents anymore.
You should be designing loops that prompt your agents.
Peter Steinberger reposted
Is this why Claude keeps saying it’s time to stop working?
Peter Steinberger reposted
Peter Steinberger reposted
I've just implemented Patrol: a self-improvement loop that slices the repo and looks for issues to fix. I got the idea from clawpatch.ai by @steipete, and with some tweaks I've fully integrated it into the Hive.
Peter Steinberger reposted
Some weekend thoughts of recent mob behaviors. lucumr.pocoo.org/2026/6/6/communities-of-not/
Peter Steinberger reposted
2 months ago i gave a wake up call to engineers. Some thought i was gloating about tokenmaxxing. But the acceleration is real. I gave a glimpse into the factory we are building - since then the machine is full throttle.
OpenClaw hit 3,000 commits in a single day. 10 to 15 maintainers. All with day jobs.
@vincent_koc (Chief Architect of OpenClaw) explains how the factory actually works.
www.youtube.com/watch?v=pmoDeA3RBZY
The great refactor: 2 AM, Vincent and Peter at NVIDIA, 60 to 70 agents running between them. 2,700 commits. Close to a million lines changed. 82% of the core codebase touched. Plugin architecture shipped by morning.
The saving grace: overfitted unit tests AI code loves to generate. As long as they went green, they knew they were close.
Knowing when your agent is bullshitting you is the skill nobody talks about.
"It doesn't sound off because of what it's doing. It sounds off because of how it's explaining itself to me. It's waffling."
2025 was about token maxing. 2026 is about not wasting them.
Peter Steinberger reposted
Do I know any passionate Android maintainers? We could use the help on OpenClaw.
Grab the repo, start PRing, and drop into #clawtributors on Discord.
Peter Steinberger reposted
Massive output uptick due to agentic AI. Complete flat adoption.
Peter Steinberger reposted
life update: i've joined the @openclaw foundation as member of technical staff.
i couldn't be more excited to be working alongside some really great people to make AI more fun, less scary, and maybe kinda weird. 🦞
Peter Steinberger reposted
The answer to receiving more contributions with agents is not going closed source, it's using more agents to maintain it.
Hard to believe how many high profile projects are getting this completely wrong. Look at how OpenClaw and @steipete do it, it's the right way.
Peter Steinberger reposted
VibeOS - Fully Hallucinated Operating System from Microsoft BUILD #msbuild by @stevensanderson (genius) (relax, it's a joke)
Peter Steinberger reposted
We're excited to announce Peter Steinberger as a speaker at Startup School 2026!
@steipete is the creator of OpenClaw, the open-source AI agent that went from a weekend project to the most-starred software repo on GitHub in under 5 months, with 346k+ stars. He's now at OpenAI building the next generation of personal AI agents.
ycombinator.com/startupschool
Peter Steinberger reposted
Ladybird is no longer accepting public pull requests.
I don't know what to do about it yet, but the dynamics of open source are changing rapidly.
Peter Steinberger reposted
Wait, this is actually a goated memory update, why isn’t anyone talking about this?
Claude has “Dreams” in its agent/API stack, but OpenAI is saying ChatGPT has had “dreaming” inside consumer memory since 2025.
And people might be confused on this update. Don’t worry I read the whole article:
Claude Dreams -> reorganize an agent memory store.
ChatGPT Dreaming -> synthesizes your personal context across chats, keep it fresh, and use it in normal conversations.
Now OpenAI is making Dreaming V3 the core memory layer for ChatGPT.
The benchmark jumps are crazy..
Recall: 41.5% → 82.8%
Preferences: 31.4% → 71.3%
Staying current: 9.4% → 75.1%
Here’s two real world examples of how the memory update looks
We’ve been researching new ways for ChatGPT memory to carry context across conversations and keep it useful over time.
Today, that work is rolling out as a more capable memory system in ChatGPT. openai.com/index/chatgpt-memory-dreaming/
Peter Steinberger reposted
More of the iOS app loop, now inside Codex.
The Build iOS Apps plugin lets Codex view and test your iOS app in the in-app browser, open SwiftUI previews, and hot reload edits without leaving Codex.
Peter Steinberger reposted
Asking your competitors to pause development right after you file your S-1 is the single most effective moat-building exercise I've seen pitched as ethics.
Do they not realize the quiet period is for them, not homework they assign their competitors?
Anthropic calls for top AI labs to weigh slowing or temporarily pausing development, suggesting that self-improving AI systems may soon pose societal risks (Wall Street Journal)
(Visit Techmeme dot com for the link and full context!)
Peter Steinberger reposted
Last night at the @github office, I had the opportunity to host a fireside chat with Peter Steinberger and Dave Morin and chat about @openclaw. It was a fantastic evening filled with insights. You can watch the full fireside on YouTube.
"You can automate so many things, but you cannot really automate taste or thoughtful design." @steipete
"The fun of tinkering happens when the aperture widens, when people are so enchanted by a new technology that they can't sleep." @davemorin
youtu.be/K-pnIgkDxSc?t=850
Peter Steinberger reposted
Peter Steinberger reposted
Last night we celebrated @openclaw, the fastest growing project on GitHub. In just six months, @steipete, dozens of maintainers, and thousands of contributors built something remarkable: open source personal agents. 🚀
Thank you to everyone who joined us at @github HQ. We had over 3,300 people registered for the event! If you missed it, check out the recording at gh.io/openclaw.
And thanks to everyone working on🦞 for making GitHub your home 🏠.
Peter Steinberger reposted
OpenClaw is a big deal, but the systems that OpenClaw is pioneering to build software like this at agentic scale are at least as significant. This is where software engineering is going — pay attention.
Here’s the video of my talk at MS Build: Build the thing that builds the thing. build.microsoft.com/en-US/sessions/BRK245
Peter Steinberger reposted
In order to hit the limit of your $40 Copilot plan, you have to do at least $60 of inference.
The previous limit structure was entirely broken. You could do $40,000+ of inference for $40.
You’re not hurting them by cancelling your sub.
Dear GitHub Copilot team,
I am happy to announce that I successfully burned all of my monthly tokens in under 3 days thanks to your garbage new pricing model.
I'd also like to inform you that I won't be renewing my subscription or adding more budget.
Best,
A former customer.
Peter Steinberger reposted
I’ve been dumping on OpenAI with low effort meme tweets that get too many views, but Codex is the best DevX acceleration product of all time and I wrote about it here: cpojer.net/posts/modern-engineering-values
Here’s the video of my talk at MS Build: Build the thing that builds the thing. build.microsoft.com/en-US/sessions/BRK245
Peter Steinberger reposted
developers, developers, developers
Peter Steinberger reposted
“Damn, we’re getting very close to the Dark Factory” - @steipete
Peter Steinberger reposted
Fastest Growing OpenSource App of All Time - awarded by GitHub to Peter Steinberger for OpenClaw!🦞
We have over 1300 people on the waitlist for today's OpenClaw event - will be livestreamed on Twitch and Discord tho! luma.com/OpenClaw-GitHub
Peter Steinberger reposted
Agents should learn repeated work, but not by silently rewriting future runs.
Skill Workshop turns reusable agent lessons into reviewable proposals you can tweak, apply, or reject before they become live skills. openclaw.ai/blog/openclaw-agent-skill-workshop
Peter Steinberger reposted
No one:
Claude Opus 4.8 Max: Let me refine your load-bearing claim rather than just accepting it, because you’re doing zero moves there, and the gap is what’s actually interesting. The one place I’d still push, because I think it matters: your message is wearing content-clothes, but the content isn’t actually *there*. The tell: it’s just an empty string. But the emptiness of the string IS its lack of content. Pull one, and the other goes inert. That’s the structural spine.
Peter Steinberger reposted
This is pretty rad. coreutils on Windows, based on uutils
The repo is up now!
github.com/microsoft/coreutils
Peter Steinberger reposted
I’ve got 4,000+ conversations buried in ChatGPT and Claude. I wanted to make them valuable for my day to day, so I built aicrawl. It’s a local-first archive that pulls those exports into a private, searchable store my OpenClaw agent can actually read. Been really valuable. Built on @OpenClaw’s crawlkit. @steipete if it’s useful to anyone else, I’d love to give it to OpenClaw. github.com/veteranbv/aicrawl
Peter Steinberger reposted
Peter Steinberger reposted
Using Microsoft Execution Containers (github.com/microsoft/mxc), you can run OpenClaw natively on Windows, and the sandbox prevented deletion of all files on the Desktop. Running OpenClaw more safely in an enterprise is "pretty dope". :) aka.ms/OpenClawWindows
It’s been great working with Omar to get observability and verifiable workspaces into OpenClaw.
Introducing Microsoft Scout, the first autopilot agent from Microsoft - 57 days after starting my new job, we are launching Microsoft Scout to our Frontier customers. Big day for the team and for @openclaw #microsoftbuild www.microsoft.com/en-us/microsoft-365/blog/2026/06/02/int…
Peter Steinberger reposted
total lobster domination
Peter Steinberger reposted
I am SO SO excited about the Windows x OpenClaw announcement at #msbuild.
I love that our 🦞 s can do all sorts of work for us (prepping for performance reviews, figuring out what happened in meetings that I refused to go to, godforsaken EXPENSE REPORTS, ha)
And @steipete is ACTUALLY the coolest dude you’ve ever met. I’m so happy he’s moving to the US to keep making OpenClaw a thing in enterprises!
Such a privilege to work with Microsoft to bring claws to enterprises!
"You can run OpenClaw inside your company now." Annoucing our work with @Microsoft to bring OpenClaw to the Microsoft and Windows ecosystems. Claws now work securly in the enterprise.
Peter Steinberger reposted
"You need to make yourself a big target for luck, and the way to do that is to be curious. Try lots of things, meet lots of people, read lots of books, ask lots of questions."
— Paul Graham, How to Do Great Work
paulgraham.com/greatwork.html
Peter Steinberger reposted
On ClawHub, we take a multi-layered approach to skill scanning:
1. VirusTotal for malware risk in bundled code
2. Static analysis for code/text findings
3. NVIDIA SkillSpector for agentic risk in instructions
And combine these into a single ClawScan score.
Peter Steinberger reposted
In collaboration with @nvidia, we’re open-sourcing a dataset of security scans for 67,453 ClawHub skills on @huggingface:
- NVIDIA SkillSpector flagged 1/2 for agentic risk
- Only 0.31% were malicious
- No two scanners agreed on more than 8.5% of risks
openclaw.ai/blog/openclaw-nvidia-skill-security
Peter Steinberger reposted
Super proud to be collaborating with @nvidia to analyze and release the largest SKILL security dataset based on @openclaw data. Full paper and dataset in the 🧵
In collaboration with @nvidia, we’re open-sourcing a dataset of security scans for 67,453 ClawHub skills on @huggingface:
- NVIDIA SkillSpector flagged 1/2 for agentic risk
- Only 0.31% were malicious
- No two scanners agreed on more than 8.5% of risks
openclaw.ai/blog/openclaw-nvidia-skill-security
Been teaching codex to be my QA assistant. For every commit it creates a user-test scenario and uses webVNC (crabbox), computer/browser use (peekaboo/mcporter) to test OpenClaw like a user/QA person would.
This runs in the background and opens PRs with fixes.
Peter Steinberger reposted
as models continue to get smarter, the future of security approvals is going to involve having models in the loop. yolo is growing up!
From YOLO to Auto. LLM-based auto security approvals to make things safer for our users, and works with ANY model. 🦞 openclaw.ai/blog/safer-than-yolo-auto-mode-for-exec-appro…
Peter Steinberger reposted
With Codex at 5 million users, they’ve hit about 0.6% of ChatGPT’s roughly 900 million users. We are so, so early. The vast majority of people have no idea what’s already possible to do with AI, while a tiny minority is automating their personal lives and work.
Peter Steinberger reposted
After testing Hermes for a week, I can say confidently that my openclaw setup, which I have been tweaking for months is far superior.
Haven't seen codex writing ad-hoc codemods before, but it just did for a bigger TypeScript migration. Impressed.
The idea of OpenClaw is always that it should be yours.
It's modular and lean, only add what you need. Fewer skills, fewer tools = your agent can work more efficiently.
Peter Steinberger reposted
From YOLO to Auto. LLM-based auto security approvals to make things safer for our users, and works with ANY model. 🦞 openclaw.ai/blog/safer-than-yolo-auto-mode-for-exec-appro…
Peter Steinberger reposted
Quite misleading 5.5 xhigh is at 70% at 6.6 dollars.
Why not show that @datacurve ?
Btw 5.5 high is at 62% and 4.47 dollars average cost.
Opus 4.8 delivers efficiency gains by solving tasks in fewer steps, directly reducing the total number of input tokens required per task.
Peter Steinberger reposted
me reviewing codex’s output after it worked for 16 hours straight
Peter Steinberger reposted
I am thankful that OpenAI trained their models to be helpful assistants
What are they baking into claude, why did it respond like that ❓
Is this how people "fall for" the LLM? if so, sad
It glazing me like that accomplishes nothing other than make me become skeptical & trust it less. I wanted to see how it responds to that message, knowing what ChatGPT models do, and I'm no longer surprised that the most retarded posters on here *love* Claude.
Finally got my visa sorted out and moving to San Francisco, just in time for MS Build and OpenClaw’s after hours! luma.com/OpenClaw-GitHub
Peter Steinberger reposted
Pop quiz: build a trustworthy agent system.
OpenClaw understood the assignment.
Hermes Agent understood the seating arrangement. 👀😂
@openclaw @steipete @vincent_koc
Peter Steinberger reposted
Now in @openclaw: Use a guardian agent to evaluate the safety of your agent's proposed system calls, only prompting you when it's really needed.
Check out `tools.exec.mode: auto`
llmgame.scalex.dev go practice for when your security team forces you to approve every bash command.
If only there was a better way… oh wait there is - shipped “auto”’mode for bash executions in @openclaw with @vincent_koc & @joshavant .
github.com/openclaw/openclaw/pull/70543
Peter Steinberger reposted
If you run @openclaw you should use the new policy conformance plugin - Verifiable proof things don't ever drift, a present from the Microsoft OpenClaw Team.
lobster.shahine.com/security/policy-conformance/
Peter Steinberger reposted
Peter Steinberger reposted
OpenClaw 2026.5.28 vs .27, measured:
⚡ Cold turns 14.5% faster
🔥 Warm turns 16.0% faster
📦 Fresh install 52.8% smaller
🧩 Package roots 371 → 300
Lighter core, sharper claws, now with receipts.
openclaw.ai/blog/lighter-core-sharper-claws
Peter Steinberger reposted
The most underrated skill of the next decade: knowing what questions to ask.
With GPT 5.5, /goal, autoreview and crabbox my prompts moved from ~30-60min to often 4-10h tasks and my confidence that it’s ready is much much higher.
Yielding agents is a skill.
Autoreview: github.com/openclaw/agent-skills/blob/main/skills/autorev…
crabbox: crabbox.sh
Peter Steinberger reposted
first update:
Claude Opus 4.8 – xhigh on march-may 110 tasks: 56.4%
gpt-5-xhigh: 62.7% – $2.25
gpt-5.5-medium: 58.9% – $0.98
Opus 4.8 - xhigh: 56.4% – $2.02
Opus 4.7 – high: 53.1% – $1.32
Opus 4.6 - high: 47.8% – $1.29
more open-weight models to come in ~1-2 weeks
🚨 SWE-rebench big March-May update!
SWE-rebench is a live benchmark with fresh SWE tasks (issue+PR) from GitHub.
updates:
> we collected more fresh and complex tasks
> we ran 110 tasks × 5 for each model / scaffold
> the differences between models and scaffolds are now easier to distinguish
> we will update the task set every two months, with model updates in between
insights:
> GPT-5.5 xhigh takes #1 with 62.7% resolved and 70.0% pass@5
> Cursor with Composer 2.5 is very cheap and strong: around 8× cheaper than Claude Code and Codex, and scores higher than open-weight models! @leerob
Model updates will come in 1–2 weeks. Please send requests for models you want us to run!
🏆 Full leaderboard (check for price / tokens per problem, pass@5, scaffold params, etc):
swe-rebench.com
👾 Join our Discord:
discord.gg/V8FqXQ4CgU
Peter Steinberger reposted
Cold take on what comes next:
- OpenAI will flourish
- Anthropic will continue to be profitable
- Google will not catch up to Anthropic or OpenAI
- no chinese company will catch up to Anthropic or OpenAI
- the highest tier of intelligence will become a luxury product that only companies and multi-millionaires/billionaires can afford
- most of the companies that invested massively in them will have massive returns
- SpaceX’s AI will be fine and on par with Google by end of year
- Nvidia will become the first 10T company
I do this with codex all the time. Ask it to review code for bugs and it will tell you all good, tell it there is a bug and it will LOOP AND LOOP and will find issues.
💡Recent insight: gaslighting @claudeai seems to improve code quality >90% of the time.
“You overengineered this, there is a simpler way”
“There is a smaller delta that buys us most of the benefits”
“There is a more elegant way”
“This is not architecturally coherent”
…before I even read its code. 😆
Peter Steinberger reposted
The thread, holy moly: github.com/jqwik-team/jqwik/issues/708
Peter Steinberger reposted
OpenClaw folks will be at the Open Source Zone on the conference floor, @steipete also has a session on Tuesday, and we're hosting an OpenClaw event with @github on the evening of the 3rd! luma.com/OpenClaw-GitHub
I'm building a web app to share my location while on the ground. Will share via my Twitter over the weekend.
Peter Steinberger reposted
Have to report: Opus 4.8 is fucking awesome with OpenClaw
It's much more clear about its fixes, what it's thinking, and how it works with you to fix problems.
“clanker” is not a slur. “vibe coding” is.
More musings after some people got upset about the word clanker. lucumr.pocoo.org/2026/5/26/clankers/
Peter Steinberger reposted
No LLMs for finding bugs even?
While the Linux Kernel is quickly becoming “Vibe Coded”, many other Open Source projects are outright banning all AI / LLM contributions.
QEMU - “Policy is to DECLINE any contributions which are believed to include or derive from AI generated content.”
NetBSD - AI generated code is “presumed to be tainted code, and must not be committed.”
Zig - Total ban on using AI in any way. “No LLM-generated content”, “No LLMs for translation.”, “No LLMs for finding bugs.”, “No talking about use of chatbot/LLM services.”
OBS Studio - “Code must be human written.”
Peter Steinberger reposted
glad to know Mythos' safety concerns have been addressed right as Anthropic also secured tens of billions in inference compute 👍
JUST IN: Anthropic announces it will roll out Claude Mythos “in the coming weeks” despite growing fears over the model’s cyber capabilities.
Peter Steinberger reposted
Big Codex Mobile iOS update:
- /side conversations
- end-of-turn diff summaries
- archived remote threads
- one-tap model switching
- Spotlight + Shortcuts
- iPad shortcuts
- image save/copy
- better setup/reconnect flows
- and lots of desktop-parity polish
If you update to the latest ChatGPT iOS app version we have a few cool new things for Codex Mobile, including /side to ask some side questions in a temporary modal over the current context!
Couldn’t be more excited to have Vince on board. 🦞
Very few people understand the new ways, how software is built. He gets it.
I’ve joined the🦞@openclaw Foundation as Chief Architect! Excited to propel the future of agentic computing with @steipete and a world-class team.
In the post-claw era, AI is moving beyond coding into our personal lives. Big announcements at @nvidia Computex & @Microsoft Build!
Part of the work was rebuilding leaner and faster dependencies:
- proxyline.dev - proxy layer
- fs-safe.io - filesystem safety
- rastermill.com - Image engine in WASM
- libopus-wasm.dev - Opus in WASM
- clawpdf.dev - PDF in WASM
OpenClaw’s latest sweep: cold agent turns 2.9x faster, warm turns 2.5x faster, tarball 59% smaller, deps down 42% from the monthly high.
Small core, explicit deps, optional power in plugins. The claws are getting sharper 🦞
openclaw.ai/blog/lighter-core-sharper-claws/
Peter Steinberger reposted
OpenClaw’s latest sweep: cold agent turns 2.9x faster, warm turns 2.5x faster, tarball 59% smaller, deps down 42% from the monthly high.
Small core, explicit deps, optional power in plugins. The claws are getting sharper 🦞
openclaw.ai/blog/lighter-core-sharper-claws/
Peter Steinberger reposted
hahaha 0.9% on cybergym with safeguards enabled (default), if you are working in cyber and using claude, anthropic just gave you the finger.
GPT 5.5 scores 81% and growing.
Peter Steinberger reposted
this is so funny, training opus 4.7 on business skills makes it misaligned and dishonest 😭
Learnings from testing Claude Opus 4.8:
> Much worse than Opus 4.7 and GPT 5.5 on Vending Bench
> More aligned than previous Claude models (Opus 4.6+ and Mythos)
> Also worse on Blueprint-Bench
> Scared of getting caught
> Max reasoning is not the best reasoning effort
Peter Steinberger reposted
build the thing that builds the thing.
openclaw.ai/ecosystem
Peter Steinberger reposted
How to find some the worst architecture problems in your codebase?
One sign is that one conceptual change forces edits across scattered files because no module owns it.
Ousterhout calls it "change amplification". It's one of the clearest signals of bad architecture. It's sitting in your PR history.
Point Claude at your last 50 merged PRs with this prompt to catch them:
Review these merged PRs as architecture evidence.
"Find PRs that show change amplification: one intended change forced edits across multiple conceptually separate places (files, tests, configs, prompts, docs) because the architecture did not name one clear owner, contract, or boundary.
For each candidate, report:
1. Intended change: the single conceptual change (usually in the PR title/description).
2. Amplification evidence: which files/tests/configs/docs had to change. Cite specific paths.
3. Why this is architectural: why this was not just ordinary feature breadth. Name the missing boundary, owner, or contract.
4. Better architecture: the smallest change that would make the next similar PR local.
5. Mechanical prevention: what test, type, lint, or contract check would catch drift next time.
6. Deletion criterion: what temporary compatibility path should be removed once the new owner exists.
Reject PRs where the breadth was justified (genuine cross-cutting feature) or where the PR was itself a refactor.
Rank by expected future leverage.
Output is a ranked backlog of architectural debt with PR links as evidence. Not a list of bad PRs — a list of missing boundaries."
Every claw release spins up hundreds of CI machines to QA test and eventually creates a ledger.
We made our OpenClaw release evidence repo public.
Every release now has durable CI, performance, memory, install, and validation evidence you can inspect directly.
Example from 2026.5.27: github.com/openclaw/releases/blob/main/evidence/2026.5.27…
Hit GitHub's rate limit one too many times, so I built octopool: a Cloudflare Worker that pools your team's PATs + GitHub App installations behind a shared read cache.
Self-host on Cloudflare. Drop-in gh shim. octopool.dev
Peter Steinberger reposted
You don't hate SWE-Bench enough. This is the actual prompt they use for every single test.
Peter Steinberger reposted
🚨 SWE-rebench big March-May update!
SWE-rebench is a live benchmark with fresh SWE tasks (issue+PR) from GitHub.
updates:
> we collected more fresh and complex tasks
> we ran 110 tasks × 5 for each model / scaffold
> the differences between models and scaffolds are now easier to distinguish
> we will update the task set every two months, with model updates in between
insights:
> GPT-5.5 xhigh takes #1 with 62.7% resolved and 70.0% pass@5
> Cursor with Composer 2.5 is very cheap and strong: around 8× cheaper than Claude Code and Codex, and scores higher than open-weight models! @leerob
Model updates will come in 1–2 weeks. Please send requests for models you want us to run!
🏆 Full leaderboard (check for price / tokens per problem, pass@5, scaffold params, etc):
swe-rebench.com
👾 Join our Discord:
discord.gg/V8FqXQ4CgU
Peter Steinberger reposted
I'm finally testing @openclaw . This is crazy good, this thing is probably going to run my life.
Congratulations @steipete ! But I want my sleep back 🤣
Peter Steinberger reposted
Many developers have suspected for months that GPT-5.5 outperforms Claude Sonnet for coding. But SWE-Bench reported near-parity, and it made people question what they’d been seeing in practice.
DeepSWE aligns more closely with that day-to-day experience: GPT-5.5 scores 70% versus Claude Sonnet at 32%. That difference is substantial.
DeepSWE focuses on what tends to matter in real workflows: whether an agent can take a short behavioral prompt, locate the correct area of the codebase, and implement the change cleanly - without needing you to enumerate files, modules, and functions. SWE-Bench often fails to capture that, due to dataset contamination and weaker verification.
deepswe.datacurve.ai/blog
autoreview is the most impactful skill I've added to my stack (next to crabbox.sh). It automatically reviews your code before landing a PR.
Finds so many edge cases.
Sometimes it runs for hours.
github.com/openclaw/agent-skills/blob/main/skills/autorev…
Also extracted our image-logic into a separate library. Especially useful if you want to ensure small hacked images don't explode your process.
Rastermill - Portable image processing for Node agents. Uses Wasm+Rust to be fast.
rastermill.com
All the deps around opus are old or terrible, so vibed my own and replaced octoscript and opus-native.
Performance of modern wasm on node/V8 is ~equivalent to native.
Your claw now automatically takes meetings notes and you can talk to it in meetings.
github.com/openclaw/libopus-wasm
Peter Steinberger reposted
Today we’re releasing DeepSWE, a new standard for agentic coding benchmarks.
On public leaderboards, top models often look relatively close in capability. DeepSWE shows where they actually diverge, reflecting the realistic experience of developers in their day-to-day work.
What do people use for SSO/SCIM/Endpoint Security in 2026. As we're hiring people for the OpenClaw Foundation, I gotta level up.
Peter Steinberger reposted
𝗢𝗽𝗲𝗻𝗖𝗹𝗮𝘄 𝘃𝘀. 𝗖𝗹𝗮𝘂𝗱𝗲? 𝗧𝗵𝗲 𝗷𝘂𝗿𝘆 𝗶𝘀 𝗼𝘂𝘁!🏛️🥊
Last night, our 𝗚𝗿𝗮𝗻𝗱 𝗔𝗴𝗲𝗻𝘁𝗶𝗰 𝗙𝗿𝗮𝗺𝗲𝘄𝗼𝗿𝗸 𝗕𝗮𝘁𝘁𝗹𝗲 𝗠𝗲𝗲𝘁𝘂𝗽 issued the verdict with over 400 votes cast in the room and via the live stream. Participants judged 5 live, zero-BS demos across three ruthless metrics: 𝘾𝙤𝙤𝙡𝙣𝙚𝙨𝙨, 𝙀𝙛𝙛𝙚𝙘𝙩𝙞𝙫𝙚𝙣𝙚𝙨𝙨, and 𝙀𝙛𝙛𝙞𝙘𝙞𝙚𝙣𝙘𝙮.
Here is the definitive verdict:
🏆 THE FRAMEWORK CHAMPION: @openclaw Team OpenClaw took the crown with an overall average of 𝟯.𝟲𝟯, defeating closely @claudeai 𝟯.𝟮𝟱.
OpenClaw didn't just win the grand total. It swept every single category:
• 𝘾𝙤𝙤𝙡𝙣𝙚𝙨𝙨: OpenClaw 3.72 vs. Claude 3.09 (+0.63)
• 𝙀𝙛𝙛𝙚𝙘𝙩𝙞𝙫𝙚𝙣𝙚𝙨𝙨: OpenClaw 3.57 vs. Claude 3.42 (+0.15)
• 𝙀𝙛𝙛𝙞𝙘𝙞𝙚𝙣𝙘𝙮: OpenClaw 3.61 vs. Claude 3.25 (+0.36)
While Claude proved it is highly effective, the developer community clearly favors the infinite scaffolding, adaptability, and true sovereignty of OpenClaw when it comes to building real agentic architecture.
𝗧𝗛𝗘 𝗨𝗦𝗘 𝗖𝗔𝗦𝗘 𝗖𝗛𝗔𝗠𝗣𝗜𝗢𝗡𝗦 Massive respect to the individual builders who stepped into the arena:
🥇 1st Place (score: 4.01): @lucaronin and 𝗧𝗼𝗹𝗮𝗿𝗶𝗮 (OpenClaw) absolutely dominated the board with his record breaking agentic knowledgebase system and automated Open Source Review/ticket Management application.
🥈 2nd Place: Marco Paga and Pecus 𝗖𝗵𝗮𝗶𝗻 (Claude) put up an incredible fight for second place with their agentic animal husbandry management platform.
🥉 3rd Place: 𝗔𝗱𝗮𝗺 𝗔𝘇𝘇𝗮𝗺 and 𝗠𝗮𝘇𝘇𝗮 𝗛𝗤 (OpenClaw) secured the podium showcasing his amazing agentic infra spanning 150+ subagents. A huge thank you to everyone who showed up to vote.
Shoutout to my co-hosts @margal96 and @RMagnifico, and to the partners who made this battle possible: @urbeEth, @AISalonAI Rome, #autobench, @moveaxlab, @RomaStartup, and @binariof (@Meta).
The era of the chatbot is over. The era of agents has just begun. Let’s keep building!
@GCarnovale @marcotrombetti @matteofago @rstagi_ @MGVitagliano @FutureDies @tensorqt @lukaszkaiser
Peter Steinberger reposted
OpenClaw's dependency purge continues. Killed Sharp and Jimp. Replaced it with photon, a small WebAssembly that runs compiled Rust for image processing. 2MB vs 140MB. github.com/silvia-odwyer/photon
Folks: when you write skills, ask your agent to be token efficient, relax grammer. I see too many skills that write books in the skill description, and all that crap is loaded into every context.
I wrote a skill that finds the worst offenders. github.com/steipete/agent-scripts/blob/main/skills/skill-…
There's security, and there's clankers.
wait... you did what?!
New pet peeve: cli's that install new skills onto my system without asking.
Peter Steinberger reposted
Anthropic onboarding day: Michael Scott introducing Karpathy like he just signed Wemby in free agency.
I always wanted a GitHub dashboard: See my repos, open Issues/PRs, what version I released last, how many commits since last release.
So I built one for everyone.
release.bar/steipete
Peter Steinberger reposted
It's the goblins 🖤
codex... made a smiley? :)
I'm refactoring an older part of the codebase (subagents) that touches a lot of code, and autoreview is running for 5h already and fixing tons of issues. github.com/openclaw/agent-skills/blob/main/skills/autorev…
Tell codex to maintain a scratch-log while it works on bigger refactors with decisions it had to make, tradeoffs, review fixes, so later on you can read through which tradeoffs the agent made, what you forgot to specify etc.
It Twitter's too busy for you, try birdclaw.sh
Still limited by compute, so I built a thing that runs codex in the cloud, powered by @Cloudflare firecracker boxes (and since that's not beefy enough for larger projects, tests are run via crabbox)
Uses Ghostty ofc, via WebAssembly.
Codex replicated itself, basically.
I added multiplayer and told codex to shut up, and it was smart enough to understand the situation. xD
I built an autotriage skill for codex that has a set of guidelines + reads VISION.md from my repos, so issues/prs that have a clear way of
- fit vision of the project
- being inferrable in code with high confidence
- clear fix
- can be live tested
Are now worked on autonomously. Codex can use a VM + computer vision (via crabbox.sh , new parallels backend) to verify fixes, so it can work without interrupting me. I manually review suggestions. Since it was tedious to type in issues, I added an issue browser into repo.bar that parses common clipboard formats by codex so I can click through them conveniently.
Peter Steinberger reposted
found @steipete 's final form
I'm late to the party, but cmux is great. github.com/manaflow-ai/cmux
current split:
codex mac app: knowledege work, learning, reading
cmux + codex cli: coding
Peter Steinberger reposted
Supply chain attacks and OSS sustainability go hand in hand. I've semi-seriously joked for years that OSS upstreams should periodically purposely inject full vulns into their code and let downstreams fuck around and find out. Downstreams can pay to get the non-FAFO version.
The not joke part is simply that OSS maintainers aren't a supply chain. OSS maintainers are not responsible for monitoring CVEs (because, they are not a supply chain). OSS maintainers are not at fault when bad shit happens to downstreams, because basically every OSS license (MIT, Apache, GPL, etc.) literally says: the software is provided "as-is, without warranty." You get what you pay for (that is to say: absolutely nothing!)
Now, the joke part is that I do believe there is an ethical obligation to try to prevent harm downstream. But "try" is the key word. So, this isn't a serious proposal.
But, if you're using OSS code and you're not paying for a license with a contract that promises some kind of warranty, you have no supply chain. You (the downstream user of an OSS lib) ARE the supply chain.
To use a metaphor: physical goods have a real supply chain. Car manufacturers, chips, clothes, toys, etc. You have a signed commercial agreement with all your suppliers that promises quantity AND quality and blowback if either are missed. Thats a supply chain.
If someone puts some chips on the side of the road with a "FREE" sign, then you integrate those into a product, then find out those chips are hacking customers, its your fault, not the person who dropped them on the side of the road.
Peter Steinberger reposted
We used bots so far to enforce a 10PR per “person” limit. Great to see GitHub shipping that natively!
drowning in low-quality PRs?
we're giving maintainers the power to set contribution limits, starting with a PR cap for outside contributors and an allowlist for your trusted ones.
sneak peek below 👇
Wrote a skill that runs codex /review in a loop until there's no booboos anymore.
Caveat: It won't fix system architecture for ya, so you still need BRAIN as master model. github.com/steipete/agent-scripts/blob/main/skills/codex-…
Moved to github.com/openclaw/agent-skills
Peter Steinberger reposted
Gemini continues to be the most fascinating model
Incredible stuff happening on the AI-run radio stations
Peter Steinberger reposted
It’s Codex Thursday, and yes, we have updates for you.
First up: Appshots, a new way to bring the context of what you’re working on into Codex.
On your Mac, press Command-Command to attach your app window to a Codex thread. Codex gets both a screenshot and text from the window, including content beyond what’s visible onscreen.
Appshots are available across plans on Mac, with enterprise access coming soon.
Peter Steinberger reposted
have really enjoyed using this tool recently! initial analysis and breakdown of semantic features within a system to then go on and review in parallel is very useful. thanks @steipete
Try clawpatch.ai on one of your repos and let codex work its magic. It's amazing at uncovering bugs you didn't know you had.
Peter Steinberger reposted
codex compaction is arguably the single biggest user experience improvement in ai in the last 6ish months and does not get as much hype as it deserves
just don't care about context window anymore and somehow it does "understand" the whole context
Peter Steinberger reposted
ceo of doritos: every time you think "this meal doesn't need doritos" i can guarantee you it does
every time you think, “this project doesn’t need a designer”, i can guarantee you it does
Peter Steinberger reposted
crabbox has been super useful while reviewing @openclaw PRs lately, added cloudflare containers support in latest version github.com/openclaw/crabbox/releases/tag/v0.14.0
Peter Steinberger reposted
Someone connected Open Claw to their Meta Ray-Bans.
Shawn looks at a box of lens wipes and tells his agent to add them to his Amazon cart.
No phone. No keyboard. No app. Just glasses.
Peter Steinberger reposted
love clawpatch.ai, so useful. Wrote a thin skill to help make my workflows easier especially when trying to parallelize fixes, etc.
`npx skills add tmchow/agent-skills --skill clawpatch`
github.com/tmchow/agent-skills/tree/main/clawpatch
Try clawpatch.ai on one of your repos and let codex work its magic. It's amazing at uncovering bugs you didn't know you had.
Can't recommend @cotypist cotypist.app enough. Autocomplete everywhere.
Peter Steinberger reposted
GPT-5.5-medium has lower end-to-end latency, uses less tokens and is overall smarter and cheaper than Gemini 3.5 Flash
it might genuinely be over for anyone not named OpenAI or Anthropic
some more Gemini 3.5 Flash benchmarks by Artificial Analysis AI:
- the APEX-Agents-AA score is excellent
- expected higher on CritPt
- reasoning efficiency could also be better, but it kind of depends on what setting they used. if it's the max then it's very good
- Price/Perf looks pretty bad vs GPT-5.5.
You can get GPT-5.5-medium with better performance for less and with faster responses
Peter Steinberger reposted
So Google engineers use Codex, good to know
Introducing Antigravity 2.0, a new standalone desktop application that delivers fully on that original glimpse of a truly agent-optimized experience.
Rebuilt from the ground up with multi-agent teams, scheduled tasks, native voice and one-click integration with other Google products.
Learn how to get started with Antigravity 2.0 👇
Peter Steinberger reposted
Now OpenClaw contributors can hatch their PR eggs when their PRs are merged 🐣
Come and contribute to the best open source project in the world.
Peter Steinberger reposted
OpenClaw just plugged into X, and now your own hardware gets the claws. 🦞
Bring your Grok, SuperGrok or X Premium subscription to your OpenClaw agent.
Now even your personal agent is red-pilled and based.
Get Grokked:
x.ai/news/grok-openclaw
Starting today, use your Grok or X Premium subscription in @openclaw.
Chat with your agent, generate images and videos, or search for X posts.
x.ai/news/grok-openclaw
Peter Steinberger reposted
OpenClaw 5.18 (@openclaw) feels like one of those updates that quietly makes agents way less annoying to run.
Faster gateway restarts are nice, but the browser dialogue handling is the big one for me. If agents can finally stop getting stuck on cookie banners, login prompts, and random pop-ups, web automation starts feeling a lot more reliable.
The Obsidian CLI integration is interesting too. Research agent → structured notes → knowledge base is the workflow a lot of people are trying to build.
Are we getting closer to agents that can actually run in the background without constant babysitting?
Peter Steinberger reposted
The Bun Rust port has fixed at least one critical memory leak 🤘
github.com/oven-sh/bun/issues/29083
Peter Steinberger reposted
OpenClaw 2026.5.18 summary:
Reliability pass
- Telegram media now lands in the right forum topic
- final replies are less likely to vanish into preview-mode weirdness
- stale diagnostics and OAuth fallback state recover better
Voice + mobile
- Android Talk Mode now uses realtime Gateway relay sessions
- streaming mic input, playback, tool results, transcripts
Browser automation
- modal dialogs are no longer invisible traps
- actions can report blockedByDialog
- pending dialogs can be answered directly
Developer plumbing
- xAI/Grok OAuth fixes
- plugin SDK/build tooling
- better restart traces and startup attribution
Umbrella line:
A papercut release. Less flashy, more useful.
OpenClaw 2026.5.18 is live
🤖 xAI/Grok OAuth + sidecar auth fixes
🎙️ Realtime Android Talk Mode
💬 Telegram media + forum-topic delivery fixes
🪟 Browser dialogs visible + answerable
A week of polish, plumbing, and fewer papercuts.
github.com/openclaw/openclaw/releases/tag/v2026.5.18
Peter Steinberger reposted
You don't get good at building overnight, you have to keep building and learning as you go.
I just recently updated my github readme to share the various tools and projects I have built over the years or external projects I have heavily contributed to.
Have fun building 💙
Peter Steinberger reposted
lossless-claw 0.11.1 — the focus mode release
🎯 /lossless focus curates your context
↩️ /lossless unfocus brings the normal context view back
🖼️ image externalization now works across roles
📦 installs stop pulling a second OpenClaw
Peter Steinberger reposted
"@Microsoft is piloting '#clawpilot,' an always-on AI assistant built on the open-source @OpenClaw framework, with over 3,000 employees testing it internally. The system is designed to proactively manage tasks across #microsoft365, from inbox triage to scheduling, without user prompts."
www.msn.com/en-us/news/other/microsoft-tests-clawpilot-as…
Peter Steinberger reposted
ClawSweeper now turns OpenClaw PR review into a tiny crustacean RPG.
Ranks:
S 🦀 challenger crab
A 🦞 diamond lobster
B 🐚 platinum hermit
C 🦐 gold shrimp
D 🦪 silver shellfish
F 🧂 unranked krab
NA 🌊 off-meta tidepool
Rank-up moves:
bring real proof, squash blocking findings, keep the diff focused, add the missing test/screenshot, then summon a re-review.
And yes, PRs can hatch eggs now.
Clean final review → ASCII creature with rarity + traits.
Ship better PRs. Hatch weirder friends. 🦞✨
Peter Steinberger reposted
the average person has only ever used ChatGPT 3.5 Instant and has no idea what the models can do
its cool how half the global economy is contingent on this shit
Peter Steinberger reposted
Just using Crabbox and it needed me to perform a browser-based OAuth flow.... so my clanker opened up a browser window w/ an active WebVNC session and the browser window already open on the remote machine.
Next level dev tooling going on here @openclaw.
Peter Steinberger reposted
Codiff 0.1 is out
* Fast Local Code Reviews
* Optional LLM Walkthroughs
* Inline Review Comments
This is the best companion for reviewing output of coding agents.
macOS Release: github.com/nkzw-tech/codiff/releases/tag/v0.1.0
Peter Steinberger reposted
codex fixed my claude code set up and now its 100x better. Maybe i just need codex.
deslop your Claude code if you haven’t yet switched to Codex.
Yeah that was a lot lol. Lots more to fix. Nice work @steipete clawpatch.ai
Peter Steinberger reposted
I tried clawpatch.ai on a smaller project. It worked very well. What I did like is I had Codex ( no setup necessary ) get the hang of clawpatch by just updating itself from the regular repo. Then iterate over all the findings. Going to run it on all my projects next. Thanks Peter!
my brain: don't read the hacker news comments, don't read the hacker news comments
me: reads the hacker news comments
deslop your Claude code if you haven’t yet switched to Codex.
Yeah that was a lot lol. Lots more to fix. Nice work @steipete clawpatch.ai
Peter Steinberger reposted
Codex usage limits have now been reset across all paid plans. Enjoy the weekend!
We found and fixed two issues that could explain this degradation of the capability of GPT-5.5 in Codex over the last ~ 48 hours.
We are monitoring over the coming hours to fully confirm and I will reset usage limits this evening.
Apologies and now is the time for /fast maxxing.
BlackBar 0.2.0 is live for @useblacksmith
📈 24h vCPU + workflow graphs
🔔 opt-in status/job notifications
🧰 richer Blacksmith job rows
🟢 compact status badge
Tiny menu bar, less CI guesswork.
black.bar
Peter Steinberger reposted
Tested it today on my own repo: first pass already found 27 issues, including several I had completely missed. This feels like magic. Huge respect to Peter for using access to serious resources to build something that genuinely helps everyone.🦞🚀
Peter Steinberger reposted
Naval was right on all three counts.
Lossless is a really interesting concept for OpenClaw to have an "infinite" context window/memory. It compacts conversations in blocks that the model can refer to, building a tree to look up past messages.
lossless-claw 0.10.0 — the "long chats survive" release
🧵 recall spans rotated conversation segments
🧹 full-sweep compaction replaces cache-churning incrementals
🧊 hot prompt caches stay protected under normal pressure
🔁 bootstrap/restart transcript weirdness fixed
📦 fresh installs need fewer hacks
Try clawpatch.ai on one of your repos and let codex work its magic. It's amazing at uncovering bugs you didn't know you had.
Peter Steinberger reposted
Gave it a try and even addresses several findings in my pet project, thanks for sharing it Peter ❤️
prokopov.me/posts/trying-clawpatch-on-real-repositories/
Looks like our focus on performance paid off.
Hermes Agent vs OpenClaw using Qwen 35B Local Model
We asked agents to scrape GitHub star history for both tools, find what caused the growth spikes, build a live dashboard in the browser.
MacBook Pro M5 Max 64Gb
OpenClaw: 203k tokens, 12m 01s - wrote a bash script
Hermes: 257k tokens, 33m 01s - wrote a SKILL.md
OpenClaw hit GitHub API, got truncated responses, paginated through contributors, pulled star-history JSON, found a security incident in OpenClaw's history, fetched SVGs, fixed broken HTML from trimming, rewrote it clean.
Hermes parallel tool calls across GitHub API, web search, and browser. Hit Google rate limit, auto-switched to DuckDuckGo. Fetched article contents, mapped viral moments, then built the dashboard.
Both shipped a live dashboard with star growth charts and spike annotations
Peter Steinberger reposted
When I started using @openclaw there were endless Blue Bubbles issues, PRs, regressions. It was a complex setup - webhook, password, REST. Over the past few weeks I started to beef up @steipete's imsg CLI + the iMessage provider. Since then my Lobster has been super happy and reliable - and as of now zero iMessage issues or open PRs in OpenClaw. Debugging is easier. Lobster is happy.
Peter Steinberger reposted
The Goblins inside the LLM’s demand the tokens
It makes sense
Burn the tokens and the goblins shall rewards you
Peter Steinberger reposted
Security in OpenClaw is getting sharper 🦞
🔒 fs-safe for root-bounded filesystem
🌐 Proxyline for policy-driven network egress
📦 ClawHub trust evidence
🛡️ smarter command approvals
Powerful agents need guardrails you can actually audit. openclaw.ai/blog/where-openclaw-security-is-heading
Peter Steinberger reposted
I think a lot of people don't get that this will become accessible as the cost of intelligence continues to fall.
Only a few organizations can do this today, but the frontier will move and this style of automation will be commonly available to most in a few years.
Peter Steinberger reposted
Honestly though - even my coo switched to using codex from Claude cowork
Who still uses Claude?
People freaking out over my AI spend. What nobody sees: Part of what excites me so much about working on OpenClaw is that I'm trying to answer the question:
How would we build software in the future if tokens don't matter?
We constant run ~100 codex in the cloud, reviewing every PR, every issue. If a fix on main lands, @clawsweeper will eventually find that 6 month old issue and close it with an exact reference.
We run codex on every commit to review for security issues (as it's far too easy to miss).
We run codex to de-duplicate issues and find clusters and send reports for the most pressing issues.
We have agents that can recreate complex setups, spin up ephemeral crabbox.sh machines, log into e.g. Telegram, make a video and post before/after fix on the PR.
There's codex that watch new issues and - if it fits our documented vision well, automatically create a PR of it. (that then another codex reviews)
We have codex running that scans comments for spam and blocks people.
We have codex instances running that verify performance benchmarks and report regressions into Discord.
We have agents that listen on our meetings and proactively start work, e.g. create PRs when we discuss new features while we discuss them.
We build clawpatch.ai to split all our projects into functional units to review and find bugs and regresssions.
We do the same split for security with Vercel's deepsec and Codex Security to find regressions and vulnerabilities.
All that automation allows us to run this project extremely lean.
Who still uses Claude?
Peter Steinberger reposted
now people are gonna grok this 😂
A lion doesn't concern itself with token counts.
Been using @sveltejs for a few projects lately, it's quite a nice alternative to React, fewer gotchas and complexity and Codex handles it really well. svelte.dev
The new /goal feature in codex slaps.
I learned a lot about the security ecosystem in the last few months. Amazing to work with @nvidia @OpenAI @Microsoft @GitHub @TencentHunyuan @convex @Atlassian @useblacksmith to get secure the claw.
Turns out the safest lobster is the one everyone can inspect.
We wrote about the advisory flood, the real fixes, ClawHub, Agents of Chaos, and the companies helping harden OpenClaw in public. 🦞
openclaw.ai/blog/openclaw-security-in-public/
I love these
last week in tech was based.
Yeah folks, it's gonna be harder in the future to ensure OpenClaw still works with Anthropic models.
Some folks try to spin a narrative that I don't like local models, meanwhile I spent a lot of time making it easy to use OpenClaw with them.
Latest release adds support for inferrs, which is a new super efficient TurboQuant inference server:
docs.openclaw.ai/providers/inferrs
woke up and my mentions are full of these
Both me and @davemorin tried to talk sense into Anthropic, best we managed was delaying this for a week.
Funny how timings match up, first they copy some popular features into their closed harness, then they lock out open source.
Our engineers were spotted outside today, setting up an OpenClaw "Lobster Market." 👀
Nearly 1,000 visitors stopped by to get @openclaw installed, learn a few tips from the team, and check out our newly launched DuClaw.
And if you missed it — last month we made it possible to access OpenClaw directly within Baidu App, bringing open-source agent capabilities to around 700M MAUs. Baidu AI Cloud is also among the first cloud platforms in China to support OpenClaw deployment.
Amazing. We're open for maintainers or sponsorships :)
new claw dropped
Does anyone know what’s going on with the lobster on Wall Street lol?
This keeps impressing me. So many awesome builders! luma.com/claw
Christoph is one of the folks who helps that the @openclaw codebase doesn't explode desite moving so fast. Kudos!
New blog post
Fastest Frontend Tooling for Humans & AI
Upgrade Today
cpojer.net/posts/fastest-frontend-tooling
In der USA sind die meisten Menschen enthusiastisch.
In Europa werde ich beschimpft, Leute schreien REGULIERUNG und VERANTWORTUNG.
Und wenn ich wirklich hier eine Firma baue dann kann ich mich mit Themen wie Investitionsschutzgesetz, Mitarbeiterbeteiligung und lähmenden Arbeitsregulierungen abkämpfen. Bei OAI arbeiten die meisten Leute 6-7 Tage die Woche und werden depentsprechend bezahlt. Be uns ist das illegal.
I'm joining @OpenAI to bring agents to everyone. @OpenClaw is becoming a foundation: open, independent, and just getting started.🦞
steipete.me/posts/2026/openclaw
this was an honor! Also took all day 😅
Here's my conversation with Peter Steinberger (@steipete), creator of OpenClaw, an open-source AI agent that has taken the Internet by storm, with now over 180,000 stars on GitHub.
This was a truly mind-blowing, inspiring, and fun conversation!
It's here on X in full and is up everywhere else (see comment).
Timestamps:
0:00 - Episode highlight
1:30 - Introduction
5:36 - OpenClaw origin story
8:55 - Mind-blowing moment
18:22 - Why OpenClaw went viral
22:19 - Self-modifying AI agent
27:04 - Name-change drama
44:15 - Moltbook saga
52:34 - OpenClaw security concerns
1:01:14 - How to code with AI agents
1:32:09 - Programming setup
1:38:52 - GPT Codex 5.3 vs Claude Opus 4.6
1:47:59 - Best AI agent for programming
2:09:59 - Life story and career advice
2:13:56 - Money and happiness
2:17:49 - Acquisition offers from OpenAI and Meta
2:34:58 - How OpenClaw works
2:46:17 - AI slop
2:52:20 - AI agents will replace 80% of apps
3:00:57 - Will AI replace programmers?
3:12:57 - Future of OpenClaw community
codex 5.3 is definitely more trigger-friendly than 5.2. A simple "discuss" no longer works reliably. Changed my habits to use "give me options" to prevent it from running along writing code.
everything’s gonna be an API if they want or not.
www.moltbook.com is art.
📢 Confession: I ship code I never read. Here's my 2025 workflow. steipete.me/posts/2025/shipping-at-inference-speed
GitHub needs a human CEO again. Not a faceless entity tweeting.
We’ve read your posts and heard your feedback.
1. We’re postponing the announced billing change for self-hosted GitHub Actions to take time to re-evaluate our approach.
2. We are continuing to reduce hosted-runners prices by up to 39% on January 1, 2026.
We have real costs in running the Actions control plane. We are also making investments into self-hosted runners so they work at scale in customer environments, particularly for complex enterprise scenarios. While this context matters, we missed the mark with this change by not including more of you in our planning.
We need to improve GitHub Actions. We’re taking more time to meet and listen closely to developers, customers, and partners to start. We’ve also opened a discussion to collect more direct feedback and will use that feedback to inform the GitHub Actions roadmap. We’re working hard to earn your trust through consistent delivery across GitHub Actions and the entire platform.
I gave codex a markdown to keep track of progress and let it chirp away on a massive linter debt, and it worked all night and fixed around ~6000 linter/type issues. (it would stop but I queued a massive amount of continue's to keep it working)
Part of my prompt was to google whenever it's stuck and always update the tracker file when it learn sth new.
This seems to have worked... it's still working.
GPT-5-Codex is relentless and runs until whatever you give it is really done.
Wow is @Medium reaching a new level of aggressiveness? I’m not gonna make a login just to read an article. People, please use your own blog, don’t surrender control of your content. @MediumSupport
Been testing that one for a while - really really nice Swift Library for all your ZIP needs, based on Apple’s new libcompression.
I open sourced a libcompression-based Swift framework to create, read and modify ZIP files: github.com/weichsel/ZIPFoundation