Armin Ronacher
@mitsuhiko.at
13K followers 220 following 1.8K posts
Creator of Flask • earendil.com ♥︎ writing and giving talks • Excited about AI • Husband and father of three • Inhabits Vienna; Liberal Spirit • “more nuanced in person” • More AI content on https://x.com/mitsuhiko More stuff: https://mitsuhiko.at/
Posts Media Videos Starter Packs
Pinned
mitsuhiko.at
Hi! In a prior life I built a lot of Python tools and libraries. I created Flask, Werkzeug, Jinja2 and many more. I worked on Sphinx and Pygments. In the years since I did a lot more Rust. For the last 10 years I spent my life working on @sentry.io.

You can find my blog here: lucumr.pocoo.org
Blog
lucumr.pocoo.org
Reposted by Armin Ronacher
gergely.pragmaticengineer.com
For an AI startup today, what language would be the best choice: Python, TypeScript, Go or Rust?

Armin Ronacher is a great person to answer: a longtime open source contributor, creator of Flask (a Python fullstack framework), and he's building his AI startup. Spoiler: he's not using Python for it.
mitsuhiko.at
I don’t let it generate tests straight away. That only comes later when I like what it did.
mitsuhiko.at
I can wrap my head around the model much better than react’s modern one. Just a much simpler thing.
mitsuhiko.at
KK. I will cover it, but it's really just: that I generate the bindings from the OpenAPI yaml. I don't spec it out, the agent creates those changes and then leverages the generated binding for the other code it writes.
mitsuhiko.at
By from spec you mean that I write the OpenAPI spec by hand? (I don't). So I'm not entirely clear what you mean here.
mitsuhiko.at
Weekend is coming up, so might take some time to share some agentic coding stuff on a video or blog post. Anything you're curious about in particular?
mitsuhiko.at
The reality is that I would not have built this project without AI. I most likely focused something else in the meantime and buy a solution for this for now. That I could build this and shape it to work exactly like I want at this stage of the project is something that AI enabled.
Reposted by Armin Ronacher
ryansolid.bsky.social
Going after OSS projects that are openly governed isn't going to do what you want it to. Vercel makes its money from its hosting platform and paid products. OSS is not a paid product and while contributes goodwill doesn't directly impact the bottomline. And these project's reach well exceed Vercel.
bnb.im
fuck vercel, fuck Next.js, fuck Svelte, fuck Turborepo, fuck SWR, fuck SWC

Posing with the architect of a genocide for sales leads is an insane and deeply immoral stance

these projects cannot separate themselves from the CEO of the company who funds them
Tweet from Guillermo Rauch, featuring a picture of him and Benjamin Netanyahu, and tweet text:

“🇺🇸 🇮🇱 🇦🇷

Enjoyed my discussion with PM Netanyahu on how AI education and literacy will keep our free societies ahead.

We spoke about AI empowering everyone to build software and the importance of ensuring it serves quality and progress.

Optimistic for peace, safety, and greatness for Israel and its neighbors.”
mitsuhiko.at
I build differently with AI. That really cannot be compared. I also don’t know how to best measure this.
mitsuhiko.at
No libs. Just execute all migrations that are not by number in my _migrations table and a pg advisory lock around it.
mitsuhiko.at
We will see in a few days :)
mitsuhiko.at
Yes, I'm majorly on codex at the moment. This is in part because I am doing more refactoring at the moment and less greenfield work, but also because the CLI got quite a bit better.
mitsuhiko.at
A few months ago there was no alternative to anthropic, today there are plenty. Shit's moving fast!
mitsuhiko.at
As for the total cost, the $100 plan for Claude is enough per month. The $25 plan gets you far for Kodaks, so those are roughly the costs per month. I think total cost, all things considered, will be around maybe $500 bucks.
mitsuhiko.at
I'm changing things regularly quite a bit. Up until recently it was 90% Claude and 10% Codex. At the moment I think it's 70% Codex, 30% Claude. Exit Strategy: Honestly, there's so many options available. It's not really a problem.
mitsuhiko.at
“Is 90% of code going to be written by AI? I don’t know. What I do know is, that for me, on this project, the answer is already yes.” lucumr.pocoo.org/2025/9/29/90...
90%
AI is writing 90% of the code I was in charge of
lucumr.pocoo.org
mitsuhiko.at
If someone wants to play around with it, the generated code is here (with preview): makerworld.com/en/makerlab/...
makerworld.com
mitsuhiko.at
My oldest and me are making a 3d printed bb8. So far we didn't make it very far, but we wanted to spray paint it, so I needed a cutout. Turns out Claude Code is great at writing OpenSCAD files. Still kinda blown away.
mitsuhiko.at
Die eigentliches Krux hier ist die Inflexibilitaet der Entstrickungsbesteuerung. Ich will das eh hier versteuern, aber ausserhalb des EWR ist sie nicht stundbar.
mitsuhiko.at
Sozusagen, ja. Das betrifft halt besonders VC finanzierte Unternehmen weil du sehr schnell einen hohen Papierwert hast. Ist aber ein Problem mit anderen Assets illiquide sind. Ohne Stiftung kann ich gar nicht aus dem Land weil es keine Moeglichkeit die Anteil in Oesterreich zu lassen.
mitsuhiko.at
Steuerliche Vorteile ergeben sich insofern, dass die Stiftung im Land bleibt und damit keine Entstrickung ausgeloest wird. Die koennte ich mir nicht leisten. Wenn ich nicht in Oesterreich gruenden wuerde, haette ich das Problem erst gar nicht.
mitsuhiko.at
Zwei Gruende: Fuer unsere Kinder, zweitens weil ich ein VC finanziertes Unternehmen habe und ich die Entstrickungsbesteuerung (Dry Tax) nicht tragen koennte wenn ich mehrere Jahre in die USA gehen muss. Ich habe dazu einfach das Geld nicht. Dazu gibt es entweder gleich ins Ausland oder Stiftung.