I ship ~10 features a day in a professional environment , with AI in the loop. At work: a Slack bot that answers questions with access to our code, tickets, PRs and logs. Another one watches monitoring alerts and only pings us when it matters. We also rewrote the Elm compiler in Rust: 176s → 12s builds. On the side: GeoQuizz, run by a team of agents for SEO, perf and QA. Here I share what actually works, numbers included.

Waiting for the real-life type benchmarks but damn, is it the first real SOTA model from Google?

There we go, the first touchscreen Macbook! And the light is back on the Apple on the back of it I don't need it, I can't afford it but damn I'd love to try it I wonder how that release will change the professional laptop market

🤤

Please get your agent environment good enough to not review the diff anymore and just review the choices

I was talking shit about Anthropic but at least, opus 5.5 has TASTE

Codex, I miss you

My GeoQuizz.io game was dropping to 6 FPS on really slow CPUs. My agents got it to 49 without changing anything visually. Turns out the globe was doing 1,920 draw calls per frame, it's now 6.

I hate creating landing pages 😩 and Sol 6.1 is SO bad

I can't handle OpenAI anymore. @claudeai I'M BACK

Ok this might be the only 500$ plan I would get

Damn, my linter rules are ahead of T3 code... that helped with the imposter syndrome

Grok bot usage is burning so fast. Where’s the @thsottiaux for @bot ?

Selected model is at capacity. Please try a different model. All day, everyday. Please @thsottiaux fix this

Switched from @Jean_build to @theo’s T3 Code last week and tweaked my workflow along the way Now merging 2x more PRs, with way less babysitting and way more confidence hitting merge

I’ll give you a better and cheaper one: use jev to get better recommendations instead of scrolling Netflix for an hour