Whenever I read discourse on AI energy/water use that focuses on the "median query," I can't help but feel misled. Coding agents like Claude Code send hundreds of longer-than-median queries every session, and I run dozens of sessions a day. On my blog: www.simonpcouch.com/blog/2026-01...

Weekend thoughts on Gas Town, Beads, slop AI browsers, and AI-generated PRs flooding overwhelmed maintainers. I don't think we're ready for our new powers we're wielding. lucumr.pocoo.org/2026/1/18/ag...

Word of the day: clauding → coding using an AI agent. "Vibe coding" has negative connotations and "AI-assisted software software development" is unwieldy. What if you use Codex? Well, if you can google using Bing, you can claude using Codex.

This is slightly depressing: Claude Code just churned 4K lines of code (addition to existing project), I just had to add two lines of config. Of course it didn't work. Spent half an hour debugging. I incorrectly copy-pasted those two lines.

AI labs scraping your sites steal your content and increase your hosting costs to boot. Agree or disagree? Here's my humble suggestion on how to solve the AI scraping problem: senkorasic.com/articles/ai-...

LLMs and other tools are great, use them wisely 🧙 fastapi.tiangolo.com/contributing...

New blog post: Don't fall into the anti-AI hype. antirez.com/news/158

StackOverflow was already in free-fall. LLMs just removed the (air) friction.

leroy jenkins maneuver

How and why I'm learning Zig: blog.senko.net/learning-zig

Straight from the Mire of the Goat, the sharpest cut of this talk I wrote 🎉 www.youtube.com/watch?v=GOrK...

Why you should self-host: blog.senko.net/developer-co...

I see a lot of complaints about untested AI slop in pull requests. Submitting those is a dereliction of duty as a software engineer: Your job is to deliver code you have proven to work simonwillison.net/2025/Dec/18/...

A 2025 vibe-coding wrap-up featuring 9 frontier models implementing Minesweeper, a silly test I used to quickly evaluate their coding: github.com/senko/vibesw... Most had no problems, so I'm retiring it as a test. Next year I'm going to be using something a bit more challenging...

We might be in an "AI bubble" thinkpiece bubble...

It's been a while since I had any trouble connecting my Linux laptop to an external display for meetings or presentations. No trouble whatsoever with printing & scanning. Even Bluetooth headphones work flawlessly. The Year of Linux on Desktop has quietly but truly arrived!

From a squirrel: "We interview devs by giving them a task to climb a tree. What is surprising: 14/15 rabbits failed this screening."

Zet CLI aplikacija. Mislim da samo mene veseli, ali neka. :)

OpenStreetMap Crash Course Održat će se u Srijedu 05.11. u 19h na: www.twitch.tv/turing_compl... Sadržaj: - uvod u OSM, kratka prezentacija s primjerima - primjer rješavanja konkretnih zadataka Trajanje cca 1h. Svi su dobrodošli naučiti malo o mapiranju. bezdomni.net/osm-crash-co...

Evo ako je itko zainteresiran malo poslušati o otvorenim kartama, pripremit ću kratko predavanje i demo cca 1h, idući tjedan jednu večer. Napravio sam poll pa se upišite kad možete. Svi dobrodošli, ne morate nužno išta raditi kasnije. :) strawpoll.com/3RnYXaxBmye

"It's hard to get someone to fix a bug when their bottom line benefits from the bug not being fixed.", to paraphrase a popular quip.

Planetary Rings xkcd.com/3156/

Succintly put: "AI is writing 90% of the code [...] At the same time, AI doesn't own the code, [I] carry the responsibility for how it runs in production." As an aside, happy to see I'm not the only maniac using AI to write raw SQL where I used to use ORM: github.com/senko/cijene...

The AlchemyConf version of my "Tell me a Story" talk is finally out 🎉 content.subvisual.com/talk-roundta...

Neka pati koga smeta, svaka bi ih majka za ZETa! (za kontekst, natječe se 25 država)

Next, people are going to call buying used books "stealing" because the authors get no money at all from the secondary sales. I better sell a few of mine that I don't like enough to keep forever, before it gets outlawed as "piracy".

Wer stoppt Siemens?

Those would simply have never gotten written before. They don't get committed. They're useful but not earth-shattering. They contain a ton of code, especially if Claude also generates tests for them. And no one's job is getting replaced by them, nor do they devalue proper software development.

When you use coding agents for something that produces the bulk of the code of an application that will be used for years, also factor in the technical debt that you are happily accumulating. When you use LLMs an an aid, you could, on the contrary, improve your coding culture.

Clearly, they are prompting it wrong.

There is a non-zero risk of AI apocalypse in which LLMs, rather than improving productivity (or at least maxxing paperclips), consume all the world's resources trying to produce anatomically and physically correct SVG of a pelican riding a bicycle. And it will all be @simonwillison.net 's fault.

Finally recorded something outside of the "studio" 😅 For this occasion I fittingly reached back to a 90s trip-hop classic by Morcheeba. Enjoy 🌊⛵🌞😎 youtu.be/FrdcxRRAibA

Recorded some updated thoughts on agentic coding tools if someone is curious. www.youtube.com/watch?v=xfm9...

The issue of sky-high expectations from GPT-5 (as highlighted by @filippo.abyssdomain.expert) is that OpenAI have been terrible at managing expectations. If you hype up "AGI achieved internally" and "PhD-level intelligence", users WILL expect miracles. Compare with Anthropic or Chinese labs comms.

"I'm just saying," the first witch said. "A cat familiar is more traditional." "Or a raven," the second witch said. "Where did she even get a dragon from, anyway?" "She tamed it. In her last job," the third witch said. "What job was that?" "Princess." "Oh. Well. No wonder she changed careers."

OpenAI released their long-promised open weight models today under clean Apache 2 licenses and with benchmarks that put them shockingly close to o3-mini and o4-mini I've run the smaller (20B) model on my Mac and it's very impressive, despite only using ~15GB of RAM simonwillison.net/2025/Aug/5/g...

Personally, I hope AI might kill (today's) copyrights and patents. It's probably one of my longest held beliefs that they are bad for innovation. At the same time such a path needs to be walked carefully and I appreciate what cloudflare is doing here. x.com/eastdakota/s...

OpenAI launched a "study mode" for ChatGPT today, and it appears to be almost entirely implemented as a system prompt Thankfully OpenAI mostly don't take measures to protect their system prompt these days so it's easy to extract it and see how it works simonwillison.net/2025/Jul/29/...

The reality of offering “unlimited” plans for any service where the provider has costs based on usage. There will always be users who abuse it in ways the provider doesn’t expect it (at least at first). For AI, generating tokens costs $$. So “unlimited” is either subsidized or removed:

I'm seeing a lot of commentary blaming the egregious data leak from the Tea dating safety app on vibe coding I'm confident that, in this particular case, that's not what happened: the code at fault looks to have been written back in late 2023 simonwillison.net/2025/Jul/26/...

This presumes "our" devices, "our" software, the services we subscribe to, work in for our benefit. In fact, they work for the benefit of shareholders. Some call it enshitification, others call it late-stage capitalism. Regardless of the name, we are mere users.

Lynx brings back fond memories of hundreds of hours spent at DEC VT110 and IBM 3151 terminals at university computing centre in Zagreb (SRCE).

I don't like podcasts (in general), as the information:banter ratio is too low for my taste. So I built myself a script that's inverse of NotebookLLM - turn any video into an article (download, transcribe, and summarize): github.com/senko/scribe

We're hiring a Head of Product at @semaphoreio.bsky.social We're not looking for a visionary. We're looking for someone who ships. semaphore.bamboohr.com/careers/38

Not using AI for programming reminds me of when we had to code C on paper in some undergrad programming courses. Yeah, you can do it, but why? I feel 2025 is the tipping point for AI use for serious software development, not just vibe coding. Here's my take on it: senkorasic.com/articles/ai-...

This snippet from simonwillison.net/2025/Jun/23/... by @simonwillison.net beautifully explains why AI-assited coding is not the end for junior devs - quite the opposite! (I'd add: "and you can ask them to explain how a particular detail works or why they've chosen this approach")

Somehow AI company CEOs have not learned that devs are VERY price sensitive and will switch without hesitation to cheaper or "free" options This is why successful dev tools startups sell brutally cheap for individual devs, then invest heavily in enterprise sales teams to make $$

I decided to record a talk on agentic programming and why I'm so damn addicted to this. I think there are a few useful things in it, but I'm sure in a month this will be outdated :) www.youtube.com/watch?v=nfOV...

The thing you have to understand is, the Enderman just has deep social anxiety. I, too, like to stare at my little favorite block and don’t like people looking at me. I’ve just been socialized to not freak the fuck out like he does. But I get it.

Download books, train AI → bad. Buy books, scan them, train AI → bad. Benj has a pretty negative bias against anything AI, but the comments section (on any AI related article) is a cesspool. Sad to see this in a site/community which was one of the best sources of tech news in the past 30 years.