🗄️

  • eng/doc.md
  • linguistics.md determinism
  • media.md print culture

🚧 dump from agentic.md

  • non-local experiments (avoid pkg installs, etc.) https://simonwillison.net/2025/Nov/6/async-code-research/
  • woodchipper https://simonwillison.net/2025/Mar/9/steve-yegge/

If you can make all of those trades, you can use agentic coding tools to produce software not merely faster than before, but better. But to do so, you need to know quite a lot about building good software already. If you've been building software poorly, agentic coding tools are just going to help you do so faster. https://davegriffith.substack.com/p/software-development-in-the-time new languages https://simonwillison.net/2025/Nov/7/llms-for-new-programming-languages/

  • https://news.ycombinator.com/item?id=44322465 https://news.ycombinator.com/item?id=44315505
  • https://colton.dev/blog/curing-your-ai-10x-engineer-imposter-syndrome/ https://simonwillison.net/2025/Oct/11/vibing-a-non-trivial-ghostty-feature/ https://simonwillison.net/2025/Oct/10/superpowers/
  • https://github.com/humanlayer/advanced-context-engineering-for-coding-agents/blob/main/ace-fca.md
  • METR study, actually not effective? https://www.astralcodexten.com/p/links-for-september-2025
  • https://www.scattered-thoughts.net/writing/everyones-got-one/
  • https://blog.florianherrengt.com/vibe-coder-career-path.html
  • garbage in garbage out

OpenAI published a very long guide to ‘prompt engineering’ for GPT-4.1. I find this kind of stuff unintentionally hilarious: if your thesis is that these models are replacing software, why do I need to memorise incantations and learn what JSON means to get the best results? All of this should be abstracted away. - Ben Evans 25.05.20

  • https://simonwillison.net/2025/Apr/20/ethan-mollick/
  • https://news.ycombinator.com/item?id=43752492
  • teaching people how to ask better questions https://paulgraham.com/writes.html

PG might be right for most people but I estimate that LLMs will actually make their users better at writing clearly. Will take the Stack Overflow ethos to a wider audience. https://stackoverflow.com/help/how-to-ask

  • Stack Overflow won in losing https://x.com/Altimor/status/1853893158368928124
  • Cursor files https://getstream.io/blog/cursor-ai-large-projects/
  • docs https://news.ycombinator.com/item?id=45884169
  • autocomplete is not the reason https://www.arguingwithalgorithms.com/posts/cursor-review.html https://ghuntley.com/stdlib/

learn how to fucking type (and how to write https://news.ycombinator.com/item?id=43739400)

  • writing https://johnstone.substack.com/p/friction-was-the-feature

Harper on An LLM Codegen Hero's Journey: "Writing skills have become critical. While we’ve always valued strong communicators on tech teams for documentation and collaboration, it’s doubly important now." One hundred thousand percent correct, I’d say. This week I gave a little talk on how to write better prompts when using Amp and I had a slide in there that said: what makes a good prompt are the same things that make a good ticket and good bug report have standards 💡️ use AI to write docs and understand the entire architecture of the system https://www.driver.ai/

  • .cursor/rules/.cursorrules https://www.nickcraux.com/blog/cursor-tip https://news.ycombinator.com/item?id=43341506 https://docs.cursor.com/context/rules-for-ai https://news.ycombinator.com/item?id=43658923

Eng leadership at my place are pushing Cursor pretty hard. It's great for banging out small tickets and improving the product incrementally kaizen-style, but it falls down with anything heavy. I think it's weakening junior engineers' reasoning and coding abilities as they become reliant on it without having lived for long, or at all, in the before times. I think may be doing the same to me too...As with so many products, it's cheap to start with, you become dependent on it, then one day it's not cheap and you're fucked.

  • what I use it for (beyond code): summarization, taxonomization, big picture, learning new concepts (regression, dimensional analysis) https://github.com/zachvalenta/apple-models-data-analysis

that little repo took me ~30 minutes total work, spanning initial idea to completion. that * ideas in a day * days in a year has been a massive delta for me. ++ I feel like my ability to use AI tooling well growing exponentially. diff btw today vs. a month ago night and day.

  • misunderstanding https://www.youtube.com/watch?v=dkV01hBdhZE https://registerspill.thorstenball.com/p/they-all-use-it https://news.ycombinator.com/item?id=41930767 makes mistakes all the time but that's why we're here! re: direnv https://chatgpt.com/c/673f8c16-e090-8004-bdc8-564bbfeb33d5

The crux of my raging hatred is not that I hate LLMs or the generative AI craze. I had my fun with Copilot before I decided that it was making me stupider - it's impressive, but not actually suitable for anything more than churning out boilerplate. Nothing wrong with that, but it did not end up being the crazy productivity booster that I thought it would be, because programming is designing and these tools aren't good enough (yet) to assist me with this seriously. https://ludic.mataroa.blog/blog/i-will-fucking-piledrive-you-if-you-mention-ai-again/ LLMs are good at explaining code. Give it code in a language you don't understand and it will explain it with 90% accuracy. https://katherinemichel.github.io/portfolio/pycon-us-2024-recap.html#simon-willison-keynote https://www.youtube.com/watch?v=ImNpR0O8nuA https://marginalrevolution.com/marginalrevolution/2025/02/o1-pro.html forcing people to think through things https://x.com/RealGeneKim/status/1853860996689064211 https://registerspill.thorstenball.com/p/surely-not-all-codes-worth-it https://registerspill.thorstenball.com/p/how-might-ai-change-programming sketching https://simonwillison.net/2024/Dec/4/steve-yegge/ https://x.com/barbell_fi https://simonwillison.net/2023/Dec/31/ai-in-2023/ https://simonwillison.net/2024/Dec/11/who-and-what-comprise-ai-skepticism/#atom-everything https://simonwillison.net/2024/Dec/10/ethan-mollick/

  • https://simonwillison.net/2025/Apr/7/john-carmack/
  • https://www.thoughtworks.com/radar/techniques/observability-2-0 https://claude.ai/chat/6bcfcae0-6294-47ef-a3bf-588a7f178c0e
  • https://crawshaw.io/blog/programming-with-llms

writing

🗄️ notes.md practice

  • have good incentives now! https://news.ycombinator.com/item?id=45623083

If your project has a robust, comprehensive and stable test suite agentic coding tools can fly with it. Without tests? Your agent might claim something works without having actually tested it at all, plus any new change could break an unrelated feature without you realizing it. Test-first development is particularly effective with agents that can iterate in a loop. https://simonwillison.net/2025/Oct/7/vibe-engineering/

  • https://www.sudowrite.com/
  • 📙 Ousterhout comments first [131]
  • https://simonwillison.net/2025/May/20/after-months-of-coding-with-llms/
  • https://bsky.app/profile/emollick.bsky.social/post/3lp5afidgvc2a

https://news.ycombinator.com/item?id=42095434 there's a fair amount of pushback as well. i align with the first comment to this guy. decent amount of pushback seems like: "im a real man, my editor is emacs, i have strong opinions about c99 vs. rust, LLMs are for wimps who write $DYNAMICALLY_TYPED_LANGUAGE_HERE" + people that are bad at writing prompts. essentially, LLMs reward the type of person who could write a good question on Stack Overflow or otherwise teaches them how to do so (if they are willing to learn) https://stackoverflow.com/help/mcve - to Josh/Kurt 24.11.15

architecture

Martin Kleppmann: “I find it exciting to think that we could just specify in a high-level, declarative way the properties that we want some piece of code to have, and then to vibe code the implementation along with a proof that it satisfies the specification. That would totally change the nature of software development: we wouldn’t even need to bother looking at the AI-generated code any more, just like we don’t bother looking at the machine code generated by a compiler.” https://registerspill.thorstenball.com/p/joy-and-curiosity-67

  • https://news.ycombinator.com/item?id=46197930

I believe the next year will show that the role of the traditional software engineer is dead. If you got into this career because you love writing lines of code, I have some bad news for you: it’s over. The machines will be writing most of the code from here on out. Although there is some artisanal stuff that will remain in the realm of hand written code, it will be deeply in the minority of what gets produced. https://registerspill.thorstenball.com/p/joy-and-curiosity-63 90% of my skills just went to zero dollars and 10% of my skills just went up 1000x. - Kent Beck https://simonwillison.net/2025/Jun/22/kent-beck/

  • https://news.ycombinator.com/item?id=42336553
  • ADRs / Stack Overflow won https://harper.blog/2025/04/17/an-llm-codegen-heros-journey/ https://registerspill.thorstenball.com/p/joy-and-curiosity-36 https://mathstodon.xyz/@tao/110601051375142142

While we’ve always valued strong communicators on tech teams for documentation and collaboration, it’s doubly important now. Not only do you need to communicate with humans, you need to write clear, precise instructions for AI. Being able to craft effective prompts is becoming as vital as writing good code. Suddenly you find yourself building out very robust specs, PRDs, and to-do docs.

  • https://steve-yegge.blogspot.com/2008/09/programmings-dirtiest-little-secret.html

For starters, non-typists are almost invisible. They don't leave a footprint in our online community...design involves communicating with other people, and design involves a persistent record of the decision tree. "And as for this non-college bullshit I got two words for that: learn to fuckin' type."

vibe coding

  • https://simonwillison.net/2025/Dec/16/gemini-thinking-trace/
  • CQ https://news.ycombinator.com/item?id=45659881
  • Github spark, Claude artifacts https://github.com/features/spark

Sparks apps are client-side apps built with React They are authenticated: users must have a GitHub account to access them, and the user’s GitHub identity is then made available to the app. They can store data! GitHub provides a persistent server-side key/value storage API.

  • https://simonwillison.net/2025/May/1/not-vibe-coding/
  • good for MVPs https://news.ycombinator.com/item?id=43576813
  • https://simonwillison.net/2025/Mar/6/vibe-coding/ https://simonwillison.net/2025/Mar/19/vibe-coding/ https://www.youtube.com/watch?v=YWwS911iLhg

10x

  • legal / medical https://www.reddit.com/r/ChatGPTPro/comments/1mt5igj/what_is_the_most_profitable_thing_you_have_done/
  • https://simonwillison.net/2024/Oct/21/claude-artifacts/
  • supercharged smart young people https://www.youtube.com/watch?v=OmiwV8kUz2E https://www.bitecode.dev/p/what-if-ai-eventually-make-programmers
  • https://marginalrevolution.com/marginalrevolution/2024/12/why-you-should-be-talking-with-gpt-about-philosophy.html

this sounds hyperbolic, but i think AI is going to be akin to the dawn of relational databases in the 80s when it comes to the software industry. re: orgs. take databases: "no you dont need all those people punching up figures. you can replace all your accountants and secretaries with cobol!". -> do we have more or less jobs centered around the production/maintenance/whatever of business data in 2024 than we did in 1974? it would seem to me a great deal more. dunno if this will hold for number of devs, but feel >=50% that there will be a fuck ton more code written in 5 years than today. bigger economy = longer tail The amount of money flowing through capitalism would astound you. The number and variety of firms participating in the economy would astound you. We don't see most of it every day for the same reason abstractions protect us from having to care about metallurgy while programming. - McKenzie https://twitter.com/patio11/status/936629780719419392 AI as a mechanical arm - you still need to know how to hit the ball, but once you do, it'll go a lot further. https://registerspill.thorstenball.com/p/joy-and-curiosity-36

you still need to know things

  • purity tests https://news.ycombinator.com/item?id=46365239
  • layman unleashes on his laptop https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code https://www.reddit.com/r/ClaudeCode/comments/1prn6nd/comment/nv3dc9i/ https://news.ycombinator.com/item?id=46080498

“When thinking about coding with LLMs, think of them as generators of templates. You say what code you need, and an LLM provides you with a template from a collection that most closely resembles the code you needed.” I’ve said something similar in different conversations these past few weeks and that I’ve begun thinking of LLMs-as-code-assistant more in the category of frameworks and generators than magic wands. https://registerspill.thorstenball.com/p/joy-and-curiosity-23

BYO

https://news.ycombinator.com/item?id=46271703 https://news.ycombinator.com/item?id=46272230 https://news.ycombinator.com/item?id=46268959

people are still not using it

I fell into an agentic coding hole earlier this year and I still haven't recovered from it. https://www.youtube.com/watch?v=tt3kY19ciFA

  • https://blog.kierangill.xyz/oversight-and-guidance
  • leave it to Kent to see it https://simonwillison.net/2025/Dec/16/kent-beck/
  • even the power users are still dim! https://news.ycombinator.com/item?id=46255285
  • https://lucumr.pocoo.org/2025/9/29/90-percent/
  • pro https://martinalderson.com/posts/has-the-cost-of-software-just-dropped-90-percent/ https://news.ycombinator.com/item?id=46196228
  • con https://andyljones.com/posts/horses.html https://news.ycombinator.com/item?id=46199723 pro https://news.ycombinator.com/item?id=46200455
  • https://simonwillison.net/2024/Dec/31/llms-in-2024/#knowledge-is-incredibly-unevenly-distributed
  • https://news.ycombinator.com/item?id=46207505
  • https://news.ycombinator.com/item?id=46227422
  • writing an algebra book https://x.com/robertghrist/status/1874105564051234951
  • Armin https://www.youtube.com/watch?v=tt3kY19ciFA