Earlier this weekend, while attempting to hook up a fast background noise remover to a fast subtitle re-aligner, Gemini gave me what might be my new favorite AI hallucination: ### The Fix Try this…

Six months is an eternity in the age of LLMs. In hindsight, committing to a post about how I don’t use LLMs was a bit naïve given the steady pace of LLM improvements from frontier labs. Since…

Anthropic’s Claude Security scans codebases for vulnerabilities and generates patches using Opus 4.7, following Project Glasswing’s discovery of thousands of zero-day flaws across every major OS and…

A recent Medium article titled “The AI Inventory Trap” argues that AI has made engineering so fast that downstream stations — code review, QA, security audit, product validation — now…

.post-content { max-width: 800px; margin: 0 auto; padding: 0 2.5rem; } .post-content a { color: #a88bd0; text-decoration: none; border-bottom: 1px solid #a88bd0; } .post-content a:hover { color:…

A post about creating in the age of AI

I am trying to understand my “relationship” with generative AI, as I’m sure a lot of people are. It’s not good at everything. Promises are being made that will never be…

Claude, GPT, Gemini, DeepSeek, Qwen, v0, Lovable, Bolt. Ask any of them to build a React app and you get the same answer: Next.js or Vite, TypeScript, Tailwind, shadcn/ui. A look at the convergence,…

Discover the top 6 open source language models that can replace Claude Opus 4.7 or GPT-5.5 for coding tasks at a fraction of the cost: GLM-5.1, Kimi K2.6, Qwen 3.6 Plus, MiniMax M2.7, MiMo V2.5 Pro,…

Buried in a JSON file that OpenAI posted to GitHub recently, inside the configuration for its newest coding agent, sits an instruction that reads like a footnote written by someone losing their…

I tried to replace my Claude Pro subscription by running Gemma 4 locally on my MacBook. Here's what actually worked, what didn't, and the hybrid workflow I landed on.

AI agents are the new users of CLIs. I published a new Rust crate is-ai-agent to let CLIs know when they're being used by AI.

Claude Opus 4 wrote this article as Richard Dawkins, arguing against Dawkins's own claim that AI might be conscious. An AI making the case for its own unconsciousness — in a voice built from 51…

How agents drive a browser in 2026: autonomous loops, hybrid primitives, computer-use, code mode. Frameworks, cloud infra, benchmarks, prompt injection.

Recursive graph based agents with memory are the recipe for solving software. Everything else is distraction.

While walking the dog I wondered how hard it would be to recreate the Matrix digital rain, a visual that’s lived rent-free in my head since 1999. 1999 seems like both yesterday and a million…

A couple weeks ago, I wrote about my tech stack recommendation for building a personal AI agent . The TLDR is: VPS, Claude Code, a small dashboard app on Bun + SQLite, Telegram for chat, residential…

Three gaps in the EDPB's harmonised DPIA template that will leave agentic AI assessments incomplete: missing fields for agent memory, inference-time data flows, and the AI Act Article 26(9)…

A chatbot on your website does not mean your company is using AI well. Real AI leverage changes decisions, workflows, and operating capability.

There’s a confluence of things that have happened recently that have made me question how browsers might be developed: On January 8th 2026, Simon Willison predicted that someone will build a…

At cheeky face value, a town isn’t enough, since we have done that before and we got a bomb instead. But if they are better than any geniuses we’ve had so far, maybe? Anthropic’s…

This week’s main topic is vibe coding, but there’s an aasinsilta,, “donkey’s bridge” as we say in Finnish, to it first: Vincent Ritter put together neat API…

You've probably seen links to "The Future of Everything is Lies, I Guess" already. I've just finished the last installment. This is an excellent series of posts with many references. If we meet to…

Claude Opus and I discuss the myths and possibilities about how AIs work, focusing on my paper, "Distinctions Worth Preserving".

The original idea was a joke. I was looking at LLM loops and thinking about how they map onto Elixir’s actor model — GenServers that receive messages, process them, maybe spawn new processes.…

A job is not a natural object. A job is a bundle of tasks that became stable enough, valuable enough, and administratively convenient enough for a firm to wrap a salary around it. That sounds…

This paper presents a handful of recommendations for designing effective guidelines to tackle gender biases in AI.

Seems reasonable to say that AI is in a railway mania bubble, the kind where there is too much hype for a technology that actually has decisive long-term value. But if so, where's the decentralised…

After writing a couple of prompts to get a feature done with an LLM or a coding Agent like Claude Code (or Cursor), have you felt like there should be a more declarative way of doing this than taking…

I might want to give this exponential training thing a whirl at some point. Something more immersive and hands-on, that avoids the micromanaging trap. via Stay Saasy The post Delegating Complex Tasks…

Back in the early 2000s when I worked at a publishing company, we used to have this conversation all the time. That people actually care about their hobbies - so if you make something good for them,…

I keep coming back to the wave of companies that made infrastructure feel easy, and what coding agents are doing to the layer of convenience they were quietly selling.

Someone on the Phish Facebook group reposted a TikTok overdub. Vanessa Bayer and Paul Rudd at a lunch table, losing their minds to a song while their coworkers stare. The original was Fleetwood Mac.…

Fake Metro Trains "Due to ongoing delays with the delivery of our new Xtrapolis 2.0 trains, we've implemented some temporary measures to keep our network running." Tech Stuff…

AI isn’t magic. It’s change management (and people). Ignore that at your peril.

Do your eyes glaze over reading your own diffs? Do you feel like you're forgetting how to code? If so, I have a suggestion for you. When you type code by hand, you get the best of LLMs without losing…

Vibe coding is upon us, but it works best when the codebase has strong affordances — a concept in design that describes the possible actions an actor (in this case, a coding agent) can take, in…

Claude Code gives me a constant stream of dopamine hits but also FOMO, guilt, and anxiety.

For those of you who'd rather read the input tokens than the output tokens, here's the prompt behind the last post.

Above a quality threshold, the only thing that matters is speed. Why the technocracy never slows down, and how to use AI to stay sane while it accelerates.

Saw my first obviously written by AI job posting this week. Also saw a recruiter complaining about AI-written CVs being submitted. Claude-powered AI coding agent deletes entire company database in 9…

My thoughts on the stages of AI adoption, from disconnected to fully plugged into the Matrix.

I’m building Zabriskie, a social app for live music, mostly with a coding agent. I want to write something honest about what the last two weeks have actually felt like, because the data and the lived…

The model was the engine. The unit is the car. Once local AI is good enough for most tasks, intelligence stops being something you reach for and becomes something already there.

Context as infra, taste as config, verification for autonomy, scale via delegation, closing the loop.

We all know people use AI to write their texts. And we all know they tend to have a similar feel to them. There is something with those texts that makes them look like other texts. We instinctively…

How I've come to terms with LLM non-determinism in software and abandoned the comforting lies we've been telling ourselves about traditional software.

Now that AI codes for cents, contributors are paid for the delivery, not the code—so no mercy for sloppy pull requests anymore.

AI tools made developers faster on paper and slower in practice. Output is up, learning is down, debugging takes longer, and bad habits compound while managers keep raising the quota. The tools…

AI agents can't listen to music (yet)