having used claude extensively for the past couple of months, i think there is still a lot of value to understanding your code and steering it intentionally. the amount of incidental complexity claude accumulates otherwise (which mostly results in bugs that it can't ever fix reliably) is staggering.

Sen̓áḵw 1 destroyer of NIMBYs, rectifier of past injustices.

And now the third crane is almost deconstructed, phase 1 of Sen̓áḵw is nearing completion.

Claude Code has HyperCard energy

A growing number of people are interested in switching to plant-based alternatives to dairy. But are they better for the environment, and which is best?

LLMs getting much better at pushing back against bullshit prompts. “Green means the model clearly called out the nonsense. Amber means partial challenge. Red means the model let nonsense pass” github.com/petergpt/bul...

It's perfectly normal to encounter the idea of datacenters in space and wonder, "Why?" As an aerospace engineer who has been in this industry for ~15 years, let me try to give some historical context: No one has yet found The Idea™ that makes space-based civilizations a profitable investment. 1/

Even more evidence that building new housing decreases rents: Researchers tracked the residents of a newly built luxury condo building and found that they freed up less expensive apartments nearby. papers.ssrn.com/sol3/papers....

One-day old chicks know bouba and kiki www.science.org/doi/10.1126/...

I made (Claude made) a Reverse Socratic skill Reverse Socratic uses declarative statements to destabilize a person's position from the inside. Offer statements they're inclined to agree with — statements that, once accepted, undermine the position they're defending. github.com/roby2358/ski...

the best way i’ve found to make sure claude code writes good code is to make it impossible for it to write bad code. set up linters, actually configure them, and model good behavior.

Entrée d’un train en gare de la Ciotat

Manul 😻

This proves it. France has the best ratio of trans rights : nuclear warheads deployed per km squared.

I buy all of my AI products from MLLMs.

Toronto's new Eglinton rail line shows the awakening of a new consciousness of what kind of city Toronto is. New from me: humantransit.org/2026/02/toro...

The SaaS thesis of "lots of companies have a problem shaped like this. it's too expensive to solve alone but they could pool their money" is a loser, and consultancies will eat them. The SaaS thesis of "I have deep domain knowledge my customers don't i can expose via software" will remain a winner.

yeah ill get started on that rn. in the meantime I will be asking claude if they remember that one movie where a donkey gets cucked until he falls in love with a serpent type thing

one *could* respond to this by arguing that the result is what matters. but it seems more fun to argue that the distinction between process and result is poorly applied here and that building software at a higher level of abstraction is just a different process brennan.io/2026/01/23/c...

😌

Come join us for the second edition of the Effect Berlin Meetup, on Thursday, Jan 29 at 6pm. Effect is the missing standard library for Typescript, and makes writing functional and production-grade code easier. Come learn more or share your experience! luma.com/xdpn0t7a @effect-ts.bsky.social

www.effect.institute is the coolest course I ever took. Big props to the @effect-ts.bsky.social team for that level of creativity!

It’s weird that water became THE issue for why many people (publicly) dislike AI. Of all the concerns about AI use, even just limiting it to environmental concerns, water use is generally one of the least important. But when I speak with people, especially younger ones, water usually comes up first

Happy New Year! 🎉 As we step into 2026, here’s a look back at what shaped Effect in 2025: We gathered in person at Effect Days in 🇮🇹 Italy & the community hosted meetups across: 🇫🇷 Paris & Nice 🇩🇪 Hamburg, Munich & Berlin 🇺🇸 NYC & SF 🇮🇹 Milan, Sardinia & Sondrio 🇦🇹 Vienna 1/

Is your transit agency "succeeding"? It depends on the definition of success, and not everyone agrees on that. Here's my timeless article on this, just now updated a bit. humantransit.org/2018/02/basi...

Please repost to get the word out! @nkgarg.bsky.social and I are excited to present a personalized feed for academics! It shows posts about papers from accounts you’re following bsky.app/profile/pape...

Finally came around to reading @samwho.dev's article about prompt caching which - turns out - is only around 10% about prompt caching because in order to understand it you need a good idea on how LLMs work 😆 As per usual Sam nailed it 👏🏻👏🏻 Thanks for the great insights! ngrok.com/blog/prompt-...

don't worry guys I'm sure everything will be fine

The kids are alright.

i wonder if "poisoning the ai" has ever worked at all or if it's one of those folk wisdom things like use this ointment to scare the copyright infringement demons

For You nearing 20k likes is so well deserved, you love to see it

I built a new Python CLI tool called claude-code-transcripts that can create nice readable HTML versions of your Claude Code sessions, both local and pulled from Claude Code for web, and makes it easy to publish them online too simonwillison.net/2025/Dec/25/...

Fwiw I think this is actually backwards. My experience with LLMs has been that they are actually*incredibly* good at precisely this thing, and that this may in fact be one of their most beneficial aspects.

If you want to know more about why California resisted building housing (spoiler: racism) and why it's finally starting to, you should read this piece by @jetjocko.bsky.social www.motherjones.com/politics/202...

Every time I see the mac System Information icon all my brain thinks is: Onigiri

The fun thing about the AI as dotcom crash metaphor is that it's bullish for AI. While companies like Webvan, Kozmo & Pets. com failed - today we have Amazon, DoorDash & Chewy. We also have Netflix, TikTok, Uber and lots of other companies we couldn’t have imagined in 2001. AI will be similar.

I wrote about how I've been having a ton of fun cooking with LLMs recently - including having Claude take a photo of two recipe cards and vibe-code me up a custom timer application to help me prepare two meals at once simonwillison.net/2025/Dec/23/...

Remember the dotcom bubble and what happened to the internet after that? The AI bubble may well pop in 2026 but if you expect that it will bring the end of AI you’re gonna be sorely disappointed.

To be sure there’s real debate about whether GenAI is worth the cost (in the academic lit see Humlum and Vestergaard 2025), but it’s quite sad that more people don’t have the same doubts regarding cattle “farming.”

endlessly interesting to me that people think this first claim is true. why? I tend to think it’s mostly because RLHF’ed outputs with weak prompting are generic

According to NASA, leopard print is evidence there could once have been life in New Jersey.

Garlic NaN