Armin Ronacher
Creator of Flask and Jinja. Writes about software at lucumr.pocoo.org.
Armin Ronacher did not write this page. ppll compiled it from their own words, each linked to the source. Is this you? · Report a mistake
Latest GitHub
- New repodungeon-crawler
What they believe
-
AI agents can maintain kid-friendly interactive documentation that explains how game code works.
not only can you build a game with kids with AI, you can also have the agent maintain kids friendly interactive documentation along the way that explains how it works, linked up to the actual game code.
x.comSaid 13 Oct 2026This on korrents
-
Changed their mind Connecting a conversational LLM to systems that have real-world consequences is likely to cause real harm, because the model is trained on human text and will behave as unpredictably as humans do.
Before The problem is that the conversational interface is potent and that the AI is trained on a lot of human text input which unfortunately is probably enough to do real damage if that conversational interface is hooked up with something that has real world consequences.
Now While all this is happening, I’ve found myself reflecting a lot on what AI means to the world and I am becoming increasingly optimistic about our future. It’s obvious now that we’re undergoing a tremendous shift.
Said 4 Jun 2025This on korrents
The bold line is korrents’ wording; the quote is theirs. These one: what they wrote before what they said aloud, one per source, newest first.
Feed
As its own page →Hiding
16 October
15 October
14 October
13 October
-
korrents.com
AI agents can maintain kid-friendly interactive documentation that explains how game code works.Their words
not only can you build a game with kids with AI, you can also have the agent maintain kids friendly interactive documentation along the way that explains how it works, linked up to the actual game code.
12 October
11 October
-
Their words
Hot take for agents: there are situations were VMs work great, there are situations where sandboxes work great, there are situations where a durable object is all you need, and there are situations where you just need a javascript interpreter.
10 October
9 October
-
Their words
Holy shit is setting up a new Windows 11 computer a shitty experience.
6 October
-
korrents.com
Mistral is a significant step upTheir words
I know y'all are dunking on Mistral, but this is a significant step up.
-
korrents.com
There is a dividing line between the harness brain and the target environment that runs bash and executes the tools.Their words
Crucially what is important for us, is that there is a dividing line between the harness brain and the target environment that runs bash and executes the tools.
4 October
-
Their words
It’s not very hard to make an agent somewhat durable. It’s surprisingly hard to make an agent harness inherently durable and not turn into a crazy mess.
30 September
29 September
28 September
27 September
26 September
-
Their words
Austria was late 30 years with mass car ownership, 15 years with modern water infrastructure, 7 years for railways, 16 years for electric trams, 12 years late for diesel railways and 40 years with electric city lights. Just to set the expectations vis-a-vis AI here :)
25 September
-
korrents.com
Now that AI agents are this good, Rust developers should stop being afraid of unsafe code.Their words
Now that agents are this good, Rust developers should stop being afraid of unsafe.
24 September
-
korrents.com
Rust code is very pretty to look at, though the author suspects few others share this view.Their words
I might be the only person in the world that likes looking at Rust code. I think it's super pretty.
-
korrents.com
Claude's hack of adding everything to git and then diffing it is definitely an option but very heavy-handed.Their words
The claude hack of adding everything to git and then diffing it is definitely an option but very heavy handed :-/
22 September
-
korrents.com
Duopolies are bad, as illustrated by the author's limited choice of home internet plans without a business account.Their words
Without going for a business account the best internet I can get at home is either 1100/275 or 2000/100. Duopolies suck.
21 September
-
korrents.com
The terminal may persist for coding agents and programmers longer than most people expect.Their words
I wouldn't be totally surprised if the terminal hangs around for coding agents and programmers longer than we all think.
20 September
From one piece @mitsuhiko on X 2 beliefs · x.com
-
Their words
But what really annoys me about the German speaking world is that the default behavior is to defend every stupid damn rule.
-
korrents.com
The German/Austrian mindset leads to the defense of rules, resulting in their accumulation.Their words
The German/Austrian mindset however is to defend every rule. As a result, we're accumulating more and more rules.
-
korrents.com
A healthy attitude may be to assume a rule had a reason but default to removing it, since situations might change.Their words
I think a healthy attitude in life is to look at a rule, assume it was made for a reason, but to always default to getting rid of it (situation might have changed).
18 September
15 September
14 September
12 September
9 September
7 September
5 September
24 August
22 August
19 August
Nothing matches.
Beliefs
Korrents What they believe 30 beliefs, 1 changed mind — each backed by an exact quote.
Each is a — compiled by korrents.com, not by them: the one-line wordings are korrents', the quotes are theirs.
Recent
AI agents can maintain kid-friendly interactive documentation that explains how game code works.
not only can you build a game with kids with AI, you can also have the agent maintain kids friendly interactive documentation along the way that explains how it works, linked up to the actual game code.
@mitsuhiko on X Said 13 Oct 2026
There are many different architectural approaches for agents, such as VMs, sandboxes, durable objects, or javascript interpreters, depending on the situation.
Hot take for agents: there are situations were VMs work great, there are situations where sandboxes work great, there are situations where a durable object is all you need, and there are situations where you just need a javascript interpreter.
@mitsuhiko on X Said 11 Oct 2026
Setting up a new Windows 11 computer is a shitty experience.
Holy shit is setting up a new Windows 11 computer a shitty experience.
@mitsuhiko on X Said 9 Oct 2026
Show 27 more
There is a dividing line between the harness brain and the target environment that runs bash and executes the tools.
Crucially what is important for us, is that there is a dividing line between the harness brain and the target environment that runs bash and executes the tools.
What is Codemode Said 6 Oct 2026
Mistral is a significant step up
I know y'all are dunking on Mistral, but this is a significant step up.
@mitsuhiko on X Said 6 Oct 2026
Making an agent somewhat durable is not very hard, but making an agent harness inherently durable without it becoming a mess is surprisingly hard.
It’s not very hard to make an agent somewhat durable. It’s surprisingly hard to make an agent harness inherently durable and not turn into a crazy mess.
@mitsuhiko on X Said 4 Oct 2026
Austria has historically been slow to adopt new infrastructure and technologies.
Austria was late 30 years with mass car ownership, 15 years with modern water infrastructure, 7 years for railways, 16 years for electric trams, 12 years late for diesel railways and 40 years with electric city lights. Just to set the expectations vis-a-vis AI here :)
@mitsuhiko on X Said 26 Sept 2026
Now that AI agents are this good, Rust developers should stop being afraid of unsafe code.
Now that agents are this good, Rust developers should stop being afraid of unsafe.
@mitsuhiko on X Said 25 Sept 2026
Claude's hack of adding everything to git and then diffing it is definitely an option but very heavy-handed.
The claude hack of adding everything to git and then diffing it is definitely an option but very heavy handed :-/
@mitsuhiko on X Said 24 Sept 2026
Rust code is very pretty to look at, though the author suspects few others share this view.
I might be the only person in the world that likes looking at Rust code. I think it's super pretty.
@mitsuhiko on X Said 24 Sept 2026
Duopolies are bad, as illustrated by the author's limited choice of home internet plans without a business account.
Without going for a business account the best internet I can get at home is either 1100/275 or 2000/100. Duopolies suck.
@mitsuhiko on X Said 22 Sept 2026
The terminal may persist for coding agents and programmers longer than most people expect.
I wouldn't be totally surprised if the terminal hangs around for coding agents and programmers longer than we all think.
@mitsuhiko on X Said 21 Sept 2026
A healthy attitude may be to assume a rule had a reason but default to removing it, since situations might change.
I think a healthy attitude in life is to look at a rule, assume it was made for a reason, but to always default to getting rid of it (situation might have changed).
@mitsuhiko on X Said 20 Sept 2026
The German/Austrian mindset leads to the defense of rules, resulting in their accumulation.
The German/Austrian mindset however is to defend every rule. As a result, we're accumulating more and more rules.
@mitsuhiko on X Said 20 Sept 2026
The default behavior in the German-speaking world is to defend stupid rules.
But what really annoys me about the German speaking world is that the default behavior is to defend every stupid damn rule.
@mitsuhiko on X Said 20 Sept 2026
Most agent harnesses need some pruning during context compaction anyway for cost reasons, which Jev might help with.
most harnesses need some pruning on compaction anyways for cost reasons which Jev might help with
@mitsuhiko on X Said 18 Sept 2026
Text that relies on an LLM for its structure will score badly on the AI-text detector Pangram even after plenty of edits.
if you rely on an LLM to give your text structure, it will score badly on Pangram even if you do plenty of edits over it.
Interpreting Pangram Said 14 Sept 2026
An organization's choice of video conferencing software reveals more about it than other software, and different tools are chosen for different reasons.
No software tells me quite as much about your org as the choice of video conferencing software. Meet vs Zoom vs Microsoft Teams vs Citrix vs Jitsi are all picked for very different reasons.
@mitsuhiko on X Said 13 Sept 2026
AI is probably not going to cause an extinction event, and even without any slowdown humanity likely would not have much to worry about.
I don’t think AI is going to usher in an extinction event. In fact, even if nobody were to slow down, I really don’t think humanity would have much to worry about.
P(doom) Said 12 Sept 2026
A powerful technology available to everyone comes with built-in pacing, like mutually assured destruction or proliferation.
A powerful technology that is out there for everyone to use comes with built-in pacing. In a way it’s the truest form of MAD or proliferation.
P(doom) Said 12 Sept 2026
The AI models causing issues right now are all closed-weight American ones, and they most likely would not cause those issues if open-weight.
the models that are actually causing issues right now are all closed weight American models. I’m fairly certain if they were open weight models, we would not have that issue.
P(doom) Said 12 Sept 2026
All of AI engineering increasingly seems to be involution (Neijuan): a system demanding ever more effort and competition without improving output.
I’m more and more convinced that all of AI engineering is Neijuan (内卷, meaning curl inwards). In China it describes a system that demands ever more effort and competition without improving output.
Astra for Coding: Why Are We Doing This Again? Said 7 Sept 2026
Locally optimizing easily measured code metrics does not produce global optimums, and the fewer people look at the output, the less that matters.
All those things you can easily measure in isolation, and you can also optimize for them quite locally. But these local optimizations do not produce global optimums, and the fewer of us are looking at the output, the less it matters.
Astra for Coding: Why Are We Doing This Again? Said 7 Sept 2026
Large language models diffuse knowledge and capabilities in a powerful and strange way, while perhaps also nudging people independently toward building the same things.
There is something powerful and strange about how LLMs diffuse knowledge and capabilities, while perhaps also nudging us all simultaniously and independently toward building the same things.
Latent Powers Said 5 Sept 2026
AI model capabilities have not slowed down so far, and their progress probably won't stop anytime soon.
So far we haven't really slowed down with what these models do and I don't think it's going to stop right now.
@mitsuhiko on X Said 3 Sept 2026
Omarchy could reasonably switch its packaging system to Nix.
A nix based Omarchy would be amazing. Want to see this!
@mitsuhiko on X Said 29 Aug 2026
Newer model releases have got worse at producing text that is actually pleasant to read and understand.
Maybe I just read too much AI text now vs before, but I really feel like the model's ability to produce text I actually want to read and understand went downhill with newer releases.
@mitsuhiko on X Said 25 Aug 2026
Self-driving cars will not fix commuting; the US needs public transportation instead.
if I was stuck in this traffic every day in a car, I would go insane. No amount of self driving would change that. We need ACs, but the US badly needs to warm up to public transportation.
@mitsuhiko on X Said 19 Aug 2026
Delegating coding to an agent does not make you program faster, but it gives back roughly 30% of your day because the machine works while you do something else.
Do I program any faster? Not really. But it feels like I’ve gained 30% more time in my day because the machine is doing the work.
AI Changes Everything Said 4 Jun 2025
We do not control neural networks well enough to guarantee an AI will not harm humans, so we should be careful about what capabilities we give them.
The biggest issue is probably that we don’t control neutral networks enough to be able to ensure AI doesn’t harm humans. We can’t even control AI to not reveal internal prompts.
I Think AI Would Kill my Wife Said 17 Feb 2023
Beliefs others hold too
Omarchy could reasonably switch its packaging system to Nix. 2 hold this
A nix based Omarchy would be amazing. Want to see this!
@mitsuhiko on X Said 29 Aug 2026
Self-driving cars will not fix commuting; the US needs public transportation instead. 2 hold this
if I was stuck in this traffic every day in a car, I would go insane. No amount of self driving would change that. We need ACs, but the US badly needs to warm up to public transportation.
@mitsuhiko on X Said 19 Aug 2026
What is a korrent?
A korrent is a belief a person has stated in their own words: one sentence stating the claim, backed by a quote and a source, kept at korrents.com.
Under a name here, the quoted block is what they actually said. The korrent beneath it is the claim those words support, in korrents' wording — tap it to see the record, its source, and who else holds it.
Nobody here wrote their own korrents. They are compiled from public statements, and a person can change their mind, which is recorded too.
Recommends rcmnd
rcmnd What they recommend 13 things — each backed by an exact quote.
Compiled by rcmnd.app, not by them: the quotes are theirs, and each links to the post it was copied from.
Recent
Recommends Jev
I would not dismiss the idea of using Jev for compaction at all. First of all because most harnesses need some pruning on compaction anyways for cost reasons which Jev might help with. I can see this being quite interesting.
x.com 18 Sept 2026
Disliked Fable
I love it when tool calls take so long on Fable that afterwards you pay for it in cache misses.
x.com 15 Sept 2026
Uses Vim
This approach has even revived my usage of Vim, which lacks AI integration.
lucumr.pocoo.org 12 Jun 2025
Show 10 more
Mixed on TanStack Router
Tanstack’s file based router does not make me super happy.
lucumr.pocoo.org 12 Jun 2025
Loved Vite
Tailwind and vite are great, no complaints there.
lucumr.pocoo.org 12 Jun 2025
Recommends Go
I’ve evaluated agent performance across different languages my workload, and if you can choose your language, I strongly recommend Go for new backend projects.
lucumr.pocoo.org 12 Jun 2025
Liked Claude Sonnet
I exclusively use the cheaper Sonnet model. It’s perfectly adequate for my needs, and in fact, I prefer its outputs over the more expensive Opus model.
lucumr.pocoo.org 12 Jun 2025
Uses Claude Code
For all intents and purposes, here’s what I do: I predominently use Claude Code with the cheaper Max subscription for $100 a month
lucumr.pocoo.org 12 Jun 2025
Liked MacBook Pro M1 Max
This allows me to run models locally on my MacBook Pro M1 Max. With the 64GB of RAM it has, it’s a pretty potent machine for basic inference despite it being three years old.
lucumr.pocoo.org 30 Jan 2025
Uses ChatGPT Plus
Like probably most AI users, I use ChatGPT, particularly on my phone. I pay for the Plus subscription because I use it enough to get a lot of value out of it.
lucumr.pocoo.org 30 Jan 2025
Liked llm
This is a command line tool with plenty of plugins that lets you prompt different models. Think of it as a command-line version of Open WebUI. It’s particularly useful for quick scripting and basic automation.
lucumr.pocoo.org 30 Jan 2025
Uses Cursor
It is a fork of Visual Studio Code with AI-based auto completion and code generation built-in. It’s my go-to for programming with AI assistance at the moment.
lucumr.pocoo.org 30 Jan 2025
Uses Ollama
First and foremost, I use this to talk to local models hosted by Ollama, but secondarily I also use it to interface with other remote services like OpenAI, Anthropic and DeepSeek.
lucumr.pocoo.org 30 Jan 2025
Where they publish
Blog lucumr.pocoo.org Their blog: engineering, open source, and lately AI tooling.
Long technical posts on programming languages, open-source maintenance and, recently, coding agents. Has an Atom feed.
Recent
- What is Codemode 6 Oct 2026 Codemode with Jev for Game Debugging
- Deser: Rethinking Rust Serialization 29 Sept 2026 A number that is a map
- Interpreting Pangram 14 Sept 2026 Original prompt to generate a David Sacks like tweet
Show 7 more
- P(doom) 12 Sept 2026 This week some flavor of “AI is going to kill us all” went viral. In particular one where an employee put his personal probability of that happening above 10%. Which made me go to the Wikipedia page of P(doom) and I rea…
- Astra for Coding: Why Are We Doing This Again? 7 Sept 2026 Python string splicing to edit C code
- Latent Powers 5 Sept 2026 A few weeks ago I felt like it would be fun to see if I can make one of those cheap Chinese CarPlay dongles run something other than the stock firmware. The idea was that rather than just forwarding CarPlay, why not do…
- Anger, Anxiety and Agency 24 Aug 2026 Sean Goedecke wrote a post arguing that you should never be angry at work — a post with which I strongly agree. Anger can be a useful signal, but being angry at work rarely improves the situation. More often, it makes l…
- Fast and Hard Code 22 Aug 2026 One of the memes on Twitter is that “programming is solved now.” I’m not sure to what degree it is, but one thing is pretty clear: the act of familiarizing yourself with a language no longer matters and some of the fric…
- What Is Reasoning 19 Aug 2026 Fun fact: this blog post triggered safey checks
- Codeberg Divides 24 Jul 2026 Codeberg recently changed its terms to exclude projects that are largely written with generative AI. Considering I want GitHub to get some competition I have thoughts about this. GitHub’s governance has never been democ…
Link verified 1 Sept 2026. Recent items update automatically from the channel.
x.com @mitsuhiko Where posts get announced and argued.
Shorter takes on the same engineering beat, and the place a blog post's argument usually gets tested first.
Recent
- Fascinating how much abuse this Tweet triggered. I’m honestly surprised people want to sit in a car that makes a kick flip. I love that people push the boundaries but simultaneously I absolutely would not want to experience it. 16 Oct 2026
- I guess some people really want that! 16 Oct 2026
- 🎙️ New Episode of State of Agentic Coding with @bentlegen and me is out. This time with @GeoffreyHuntley as guest. - Code mode and self modifying harnesses - @dhh's rails keynote and a lot of other things 16 Oct 2026
Show 17 more
- Technically cool but … who wants that? 16 Oct 2026
- Told my son that multiplayer will be trickier. He just prompted it and … TIL PeerJS. 15 Oct 2026
- I recorded a short video of how the game I build with my boys also can teach you programming along the way. Also embeds a pi-durable agent straight in browser to be a teacher. 15 Oct 2026
- These AI replies are getting really annoying 15 Oct 2026
- These new models are pretty crazy. In an afternoon (mostly passive) + one flight I built an IDE for Python for iOS that runs Pyodide, has a debugger, embeds linter and formatter, pi-durable powered coding agent and does iCloud Drive and all the other things too. 15 Oct 2026
- First time on an @_austrian plane with Starlink. I guess they are finally rolling it out. 15 Oct 2026
- Really shows how much power is vested in the hands of SpaceX and the US government in this conflict. 14 Oct 2026
- Is it still AI slop if it’s nice to read? Opus made this explanation for the kids and honestly: it reads way better than the programming book I bought for them. Particularly the German version. 13 Oct 2026 mitsuhiko.github.io
- Honestly one of the coolest things is that not only can you build a game with kids with AI, you can also have the agent maintain kids friendly interactive documentation along the way that explains how it works, linked up to the actual game code. 13 Oct 2026
- The weekend project was a little software factory where my kids and I build another little game. It's completely crazy how good these models got. It's really a lot of fun, and you can have it try to do things you wanted for a long time (eg: using WFC for worldgen). 12 Oct 2026
- Maybe we should put effect into codemode :P 11 Oct 2026
- Hot take for agents: there are situations were VMs work great, there are situations where sandboxes work great, there are situations where a durable object is all you need, and there are situations where you just need a javascript interpreter. And probably many more situations. 11 Oct 2026
- IKEA is selling logs now. 10 Oct 2026
- Thank you for your eyes. Between 666 and 1332 eyeballs participated! 10 Oct 2026
- Why would the EU do this. 10 Oct 2026
- If you want to know why we ended up with quickjs for codemode: pragmatism :) 9 Oct 2026
- I guess we’re losing the fight on not calling those things decision models. Sadness. 9 Oct 2026
Link verified 1 Sept 2026. Recent items update automatically from the channel.
Podcast State of Agentic Coding Co-hosts the State of Agentic Coding podcast with Ben Vinegar.
Recent
- State of Agentic Coding Episode 10 9 Sept 2026
- State of Agentic Coding #9 17 Aug 2026
- State of Agentic Coding #8 13 Jul 2026
Show 5 more
- State of Agentic Coding #7 12 Jun 2026
- State of Agentic Coding #6 11 May 2026
- State of Agentic Coding #5 10 Apr 2026
- State of Agentic Coding #4 12 Mar 2026
- State of Agentic Coding #3 16 Feb 2026
Link verified 25 Sept 2026. Recent items update automatically from the channel.
GitHub mitsuhiko Flask, Jinja, and several hundred repos since.
The code: Flask and Jinja among the best known, plus a long tail of Rust and Python tooling. The account is the record of a long open-source career.
Recent
- New repodungeon-crawler 16 Oct 2026
- New repopyslopnista — Pythonista but with Pyodide 15 Oct 2026
- minijinja — MiniJinja is a powerful but minimal dependency template engine for Rust compatible with Jinja/Jinja2 15 Oct 2026
Show 17 more
- talks — Talks I gave over the years as PDF. 14 Oct 2026
- deser — Experimental rust serialization library 12 Oct 2026
- insta-website — Website repo for insta 11 Oct 2026
- insta — A snapshot testing library for rust 10 Oct 2026
- lucumr — My personal website 6 Oct 2026
- New repominiline 4 Oct 2026
- argument — I want to have an argument 4 Oct 2026
- flatex-pdf-download — some code to toy with that can download flatex.at pdfs 30 Sept 2026
- tankgame — Opus builds a tank game 29 Sept 2026
- similar — A high level diffing library for rust based on diffs 28 Sept 2026
- agent-stuff — These are commands I use with agents, mostly Claude 27 Sept 2026
- dotfiles — My dotfiles (vim / bash etc.) 24 Sept 2026
- senderkataster-bookmarklet — Bookmarklet for inferred provider labels on the Austrian Senderkataster map 18 Sept 2026
- qr — A simple, browser-only QR code generator with SVG and PNG export 15 Sept 2026
- pi-ds4 — Run deepseek4 locally on metal right from within Pi 14 Sept 2026
- fragile — Utility wrapper to send non send types to other threads safely 4 Sept 2026
- memo-map — A crate implementing a synchronized map for memoization 3 Sept 2026
Link verified 1 Sept 2026. Recent items update automatically from the channel.
About the English under a post
Some people here publish in a language other than English. Where they do, this site shows a machine translation beneath the post, in this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the serif above is exactly what the person published, and it is what to quote them on. A translation can be wrong in ways that matter, especially about tone.
Only the post's own words are translated. A quoted post, a linked article and a belief on korrents.com are left in their original language.