Thomas Witt (@thomas_witt) header
Thomas Witt (@thomas_witt) profile picture

🚀 Founder & CEO @ https://teapawt.net/t.co/2vEoCACCZP •🪽Angel Investor @ https://teapawt.net/t.co/L7uzPUuzSb

Joined November 2007
Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)Photo by Thomas Witt (@thomas_witt)
Thomas Witt retweeted
Met a German founder this week and asked him if all the stories one reads about the challenges of startups in Germany are exaggerated. "No, they're understated." Proceeded to describe spending a full day having a 90-page investment contract read to him (mandatory under German law; § 13 BeurkG) by a notary that then charged €30,000. That was for his first company. His second company, needless to say, was not incorporated in Germany.
1,042
2,141
655
29,885
12,138,605
Running Claude Code on Fable or Opus 5 and wondering where your tokens go? Might be your subagents. These few lines of code can save you a lot of tokens, because they remind Claude Code to use the model you actually chose for your subagents: thomas-witt.com/blog/blog-su…
1
61
Ghostty + herdr = Agentlove 🤖💚 I switched from iTerm2+tmux/zellij - and it fits my workflow perfectly - and is much fast on my Mac. Can only highly recommend to look into it.
1
66
I recently came across crabbox.sh. Parallel coding agents were eating too much Mac resources long integration tests. It boots a custom AMI on an AWS spot instance with the full DB stack, runs tests there, falls back to local if AWS is down. An env var forces remote.
3
78
Thomas Witt retweeted
We’ve received notice that the Department of Commerce has lifted export controls on Claude Fable 5 and Mythos 5. We'll begin restoring access tomorrow, and will share an update soon. We’re grateful to our users for their patience, and to everyone who worked with us on redeploying the models.
3,989
12,581
7,034
83,479
15,197,243
Thomas Witt retweeted
Replying to @sabinedoering
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.
530
1,061
398
10,823
2,349,674
Thomas Witt retweeted
I think people are sleeping a bit on how much Ruby on Rails + Claude Code is a *crazy unlock* - I mean Rails was designed for people who love syntactic sugar, and LLMs are sugar fiends.
260
201
161
2,401
687,200
$285B wiped out because Anthropic posted a text file on GitHub. Not a new model.Just ~2,500 lines describing how to review contracts—first-year law school stuff. "Sophisticated" traders panic-sold on a headline without checking anything. Traders have no idea what they trade on
188
Finally codex supports conversation steering as experimental feature and is catching up to claude code. Also worth reading regarding codex as agent: developers.openai.com/codex/…
1
85
“Read the headlines: AI will make developers obsolete. The era of the developer is over. When cloud computing arrived, they feared automation would make them obsolete. Instead, it created an explosion of new companies and new engineering roles” – @Werner allthingsdistributed.com/202…
43
Some examples of people quickly accomplishing ambitious things together: patrickcollison.com/fast
34
The latest 2.0.64 release of Claude code is interesting. Besides finally named sessions and async agents, they support so-called rules. Maybe that's finally a solution that claude follows coding styles better: code.claude.com/docs/en/memo… (github.com/anthropics/claude…)
122
First tests with @OpenAI 's new gpt-5.1-codex-max - even on xhigh - are very disappointing. It doesnt follow instructions well and I can see nowhere that its working 24 hours on tasks - it basically gives immediately answers to very complex tasks, which can't be right. #codex
1
1
76
gpt-5.1-codex gives for me (Ruby on Rails codebase) MUCH better results on the high setting.
57
Coding Agents Tip: When implementing large features, you MUST write a markdown spec doc first, divided in phases & subtasks (#Claude in ultrathink plan mode is quite good at this). Then ask #codex to implement it and stop at each phase. Then run claude & another codex to review.
54