Pierre Chapuis
catwell.info
Pierre Chapuis
@catwell.info
https://catwell.info

Software @ https://finegrain.ai/. AI and Distributed Systems.

Known as @pchapuis or @catwell elsewhere.
For those who are already using coding agents 100% of the time or close, this is important: steve-yegge.medium.com/the-ai-vampi...

The "Claude as a pair" way to work is by far my favorite but it is not sustainable because of this. As models and agents improve we'll need to learn to really delegate.
The AI Vampire
This was an unusually hard post to write, because it flies in the face of everything else going on.
steve-yegge.medium.com
February 12, 2026 at 7:53 AM
Vibe-coded project of the day. Watch logs from modal runs & kill them.

Made with Teal this time, because agents can use languages they don't know much too. I have a real use-case for this, but mostly it was to test terminal.lua.

github.com/catwell/mult...
GitHub - catwell/multimodal: A CLI tool to display the output of several Modal jobs.
A CLI tool to display the output of several Modal jobs. - catwell/multimodal
github.com
February 9, 2026 at 9:50 PM
@keltounet.keltia.net Je viens de voir ton update sur blog.keltia.net/la-maison-sa...

Ayant longtemps habité pas loin (rue du Dôme) je suis deçu aussi. J'ai immédiatement regardé ce qu'il advenait de Shiki et je suis rassuré : les cuisiniers de Sanki les ont rejoints shiki-boulogne-billancourt.com
La maison Sanki à Boulogne
Suite un déplacement un peu inattendu du côté de Boulogne-Billancourt — un besoin de convoyer mon ainée au spectacle d’une de ses copines de classe — je suis retourné dans un des meilleurs restaurants...
blog.keltia.net
February 8, 2026 at 12:13 PM
At each of the companies I worked for, I have written a blog post describing what I was doing there after a few years. In keeping with the tradition, here it is for Finegrain: blog.separateconcerns.com/2026-02-07-f...
What I do at Finegrain
What making lightweight image editing models that run on edge devices entails.
blog.separateconcerns.com
February 7, 2026 at 2:18 PM
Several of the best engineers I know have joined the performance teams of Anthropic and OpenAI. They are probably part of the few teams with the highest concentration of talent in the world by now.
February 7, 2026 at 8:46 AM
Back to training -> lament existing tooling -> vibe-coded what I lack.

If you use Trackio for experiment tracking with concurrent remote (serverless) runs you may like this.

github.com/finegrain-ai...
GitHub - finegrain-ai/trackio-tool: A tool to work with Trackio data files.
A tool to work with Trackio data files. Contribute to finegrain-ai/trackio-tool development by creating an account on GitHub.
github.com
February 6, 2026 at 5:34 PM
I am training models again. It has been a long time 😅
February 3, 2026 at 6:20 PM
Reposted by Pierre Chapuis
I wrote this in April, before Claude Code and Codex. Now I am changing my mind: The Carpenter can indeed just ask for a nice table.
January 28, 2026 at 7:11 PM
JB Kempf highlights a key benefit of coding agents in this French podcast: the ability to greatly accelerate feature bootstrapping.

www.youtube.com/watch?v=25Mw...
Vibecoding : Évolution ou Révolution ?
YouTube video by A la french 💻
www.youtube.com
January 28, 2026 at 12:50 PM
github.com/badlogic/pi-... Oh, I have a new harness to try!
Add OpenCode Zen provider support · badlogic/pi-mono@97d0189
github.com
January 9, 2026 at 10:51 PM
If you have not tried Opus 4.5 because of the cost yet, you have no excuse anymore. ampcode.com/news/amp-fre...
The Frontier Is Now Free
$10/day for any mode, including smart with Opus 4.5
ampcode.com
January 9, 2026 at 5:15 PM
I had Gemini do this with Nano Banana and it was not as straightforward as I hoped it would be...
January 8, 2026 at 9:55 PM
Companies losing revenue because people use LLMs instead of the Web is no longer theoretical: github.com/tailwindlabs...

The whole advertising model needs to be rethought.
feat: add llms.txt endpoint for LLM-optimized documentation by quantizor · Pull Request #2388 · tailwindlabs/tailwindcss.com
Add /llms.txt endpoint that serves a concatenated, text-only version of all Tailwind CSS documentation pages optimized for Large Language Model consumption. Extract text from MDX files, removing J...
github.com
January 8, 2026 at 10:42 AM
Currently trying OpenCode Zen with free GLM 4.7 (opencode.ai/docs/zen/), it works super well.
opencode.ai
January 2, 2026 at 1:13 PM
Probablement l'auteur de SF francophone que j'ai le plus apprécié. RIP.
Message pour tous les lecteurs et lectrices...

On vient d'apprendre avec une grande tristesse la mort de Pierre Bordage, immense auteur d'imaginaire (et pas que).

buff.ly/zrMXKrg

On a une grande pensée pour ses proches...

#sciencefiction #fantasy #PierreBordage
December 30, 2025 at 8:21 PM
The thing I like most with coding agents is they let me have side projects again. It has been years since I could work on anything significant, because kids. Last week I could test *two* long-time ideas in a handful of hours each.
December 29, 2025 at 3:09 PM
Reposted by Pierre Chapuis
🎄 Before heading to holidays, I wrote my 2025 year in review.

✍️ Back to engineering after 2 years of management. Built moonpool, a #FoundationDB simulation framework in #rustlang. Published 20 posts. Learned to work with LLMs. A year of going deeper.

pierrezemb.fr/posts/2025-y...
2025: A Year in Review
Reflections on returning to engineering, discovering simulation as a superpower, learning to work with LLMs, and the compounding value of years of distributed systems work.
pierrezemb.fr
December 26, 2025 at 10:01 AM
Winter update about how I use coding agents. tl;dr Claude Code all the things. blog.separateconcerns.com/2025-12-26-a...
How I use AI coding tools (in winter 2025)
Claude Code all the things.
blog.separateconcerns.com
December 26, 2025 at 11:27 AM
I went back to playing with Zig after a while and here is something that took me some time to figure out: if there is a symlink in the path of your global cache, you will have issues (e.g. I symlink ~/.cache). Set ZIG_LOCAL_CACHE_DIR to the real path.
December 23, 2025 at 12:58 PM
L'Allemagne et l'Italie étant les deux autres gros budgets de la défense Européenne chère à Macron, je propose de nommer le nouveau porte-avion de personnalités historiques en leur honneur.

Le PANG Dante - Hegel.
December 22, 2025 at 7:09 PM
Reposted by Pierre Chapuis
En ce moment je contacte des gens qui ont un blog pour comprendre leur process d'écriture, plateforme idéale etc...

Je construis une alternative européenne à medium, substack etc... Si ca vous dit de tester, ou juste discuter de ce que vous aimeriez trouver ou pas ne pas trouver. N'hésitez pas :)
December 21, 2025 at 2:02 PM
😢
Nous venons d'apprendre la disparition de @bruno.treguier.org 😭 Toutes mes pensées vont à la famille 🫶
December 19, 2025 at 5:26 PM
The Arch Linux WSL image is now reproductible

antiz.fr/blog/the-arc...

Reproductibility is hard, important work that has been going on for years. It's so cool that we're now reaching milestones such as this!
The Arch Linux WSL image is now reproducible
I’m happy to share a great milestone I worked on and that we’ve been able to recently reach: The Arch Linux WSL image is now bit-for-bit reproducible! I attended this year’s Reproducible Builds Summit...
antiz.fr
December 18, 2025 at 9:08 AM
> If you keep a menagerie of “too hard for AI” projects, then you will be able to watch and measure their cognitive progress increasing month by month.

steve-yegge.medium.com/six-new-tips...
Six New Tips for Better Coding With Agents
I’m hanging out in Sydney with my esteemed co-author and co-conspirator Gene Kim today; we flew in to conduct Vibe Coding workshops and…
steve-yegge.medium.com
December 13, 2025 at 12:35 PM
@tomfevrier.bsky.social since you left X: @tomforth.co.uk would like to make the same maps you did for train vs car in France but for GB. I think you live over there now so maybe you already did?

x.com/thomasforth/...
x.com
December 6, 2025 at 11:01 AM