Peter Steinberger (@steipete) header
Peter Steinberger (@steipete) profile picture

Polyagentmorous ClawFather. Came back from retirement to mess with AI and help a lobster take over the world. @OpenClaw🦞 + @OpenAI

San Francisco, CA
Joined March 2009
Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@steipete)Photo by Peter Steinberger (@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. youtube.com/watch?v=wZYnJfL4…
97
73
17
1,160
447,638
Peter Steinberger retweeted
breathtaking.
8
12
5
342
51,778
Peter Steinberger retweeted
As Apple develops its own computer-use agent, it’s making it harder for rival agents to control Macs. In macOS 27.3, enabling ChatGPT’s third-party computer-use support now requires 14 distinct steps, up from four. Apple cites security; Some users see anti-competitiveness.
NEW: Apple is developing an autonomous AI agent with computer-use capabilities, including device control, app interactions, web browsing and coding. The effort goes well beyond Siri and aims to compete with Meta’s Muse, ChatGPT’s Dots and Claude Code. bloomberg.com/news/articles/…
183
149
94
1,915
213,530
Landed a fix for that in OpenClaw yesterday (for the few folks using the embedded harness instead of codex, since it’s a plugin), most harnesses aren’t updated yet.
If your GPT-6.2 Astra or Sol API bills suddenly went up this week, there's another caching bug worth knowing about. OpenAI Codex issue #62312 found that on GPT-6.2 Astra and Sol, anything placed in the top-level instructions field of a Responses API request isn't actually cached. It stays at 0 cached tokens on every turn, even though OpenAI's own diagnostics incorrectly report a cache hit. The impact is pretty big. Moving from GPT-6.1 Sol to 6.2 Sol dropped actual cache hit rates from 97.4% to 46.4%, meaning 64K–72K system tokens were billed at full price on every step. There's already a workaround in OpenCode: put the system prompt inside a leading developer message in input.
38
9
3
241
43,239
Peter Steinberger retweeted
i cant believe people are still using coding agents in the terminal in 2026
262
26
45
981
180,909
OpenAI is a research lab pretending to be a ruthless profit-maximizing corporation, and Anthropic is the exact opposite
78
219
23
5,037
192,202
Peter Steinberger retweeted
excellent reporting by @jasminewsun in the NYT today about the strange and idiosyncratic myths that the company behind the soon-to-be largest IPO in history believes about itself
3
39
4
579
71,753
Peter Steinberger retweeted
Days like this I genuinely feel like I live in a different universe from the rest of the world. Saw ubiquitous 6.2 hate / it's over for OpenAI yada yada ... ran the highest-signal autoresearch bench and 6.2 sol/astr are *absolute monsters*. Astra 6.2 saturates it at low effort
Replying to @_arohan_
Astra 6 was better than Fable 5.2 and Opus 5.5 on my highest signal OOD autoresearch eval. Haven't tested Astra 6.2 but assume it's modestly better. I can't speak to internal models but do you really prefer their public models for research that much? nitter.cf/JoshPurtell/status/212…
61
48
5
1,262
163,153
Peter Steinberger retweeted
I just had my bot call Frontier (Fiber) and the customer support rep immediately denied my bot saying we dont talk to machines transfer me to your human. So companies can use agents on us but we cant use agents on them? Any company that denies bots will lose.
58
28
6
1,375
38,270
Peter Steinberger retweeted
Hogwarts will not reinstate Harry Potter
154
460
52
7,316
163,328
Peter Steinberger retweeted
This is our high level architecture for Microsoft Autopilot and where @openclaw fits in
20
37
2
340
33,380
Peter Steinberger retweeted
Quad9's forced withdrawal from France highlights the dangers of mandating DNS-level blocking. Read more: cfl.re/4f0cowv
44
299
50
1,749
376,194
Peter Steinberger retweeted
Not a fan of this on macOS 27.3 beta 2. Terminal command + reboot removed. sudo mkdir -p /Library/Preferences/FeatureFlags/Domain sudo defaults write /Library/Preferences/FeatureFlags/Domain/TCC AX_TRANSPARENCY_INDICATOR -dict Enabled -bool false
6
9
2
78
17,903
Peter Steinberger retweeted
repeat after me: oneshotting a threejs "game" is not indicative of real world model performance. i'm tired.
114
135
17
2,335
79,554
Peter Steinberger retweeted
Changes beyond Full Disk Access that have gone under the radar: 1. New in beta 2, *no panel* in Privacy & Security allows dragging in an app. Overlays from apps like ChatGPT won’t work anymore. Users have to use the + button to add apps. 2. As expected, TCC stops recording automatic FDA denials, so apps no longer show toggled off on the list as an easy way to enable. The TCC auto-denial policy affects: • Full Disk Access • Files & Folders (Private Folders) • System & App Protection Override (old Accessibility) The no-app-dragging change affects all Privacy & Security categories: • Accessibility • App Management • Bluetooth • Developer Tools • Files & Folders (Private Folders) • Full Disk Access • Input Monitoring • Local Network • Media & Apple Music • Screen & System Audio Recording • System & App Protection Override (old Accessibility)
26
30
22
441
147,826
Peter Steinberger retweeted
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. 🤦‍♂
Replying to @mattt
move to cloud agents. local isn't the future
26
4
206
43,246
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 macrumors.com/2026/11/16/mac…
243
67
30
1,679
154,794
Peter Steinberger retweeted
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.
162
75
22
1,503
90,371
Peter Steinberger retweeted
Codex team might have actually fixed the "model at capacity" errors that have been pestering you.
2,532
250
399
9,267
1,442,587
Peter Steinberger retweeted
I nominate this typical Londoner for the Nvidia CEO job - impossible to honeypot
124
995
63
36,849
1,098,367