Two important essays on AI in drug discovery landed in August. Both are worth your time. Both have the right intuitions. And one of them is subtly framing the wrong problem, in my opinion. The first…

Datatrove just had a new release which added support for a Jobs backend. In this blog post I’ll show how we can use this to create a dataset for fine-tuning a small task-specific model with zero…

Use Gemini 3.7 Flash Computer Use with about 150 lines of Python to control an Android emulator. The model plays Wordle from screenshots: opens Chrome, clears popups, reads tile colors, and solves in…

A longstanding grievance with several social media platforms has been how they promote the use of AI in content creation, but for all the…

The bottleneck is decisions per hour. Scaling past one human per agent means agents supervising agents, and a control surface built for machine operators.

"Good enough" no longer feels good enough

Ever tried explaining why somebody needs a 3D printer? "It prints stuff" sounds profoundly unimpressive. 3D-printer owners make it worse by enthusiastically showing you the ugliest plastic objects…

Reasoning scores climb while per-token compute drops. Labs are stripping world knowledge out of models, and I think that's the right trade.

How caching changed once LLMs entered the picture: what prompt caching actually stores.

Coding with AI means different things to different people. Some call it glorified autocomplete with Github Copilot, others copy-paste code snippets from ChatGPT, and some hand over control over their…

Coding agents are designed to help. Give them a task and they will usually start trying to solve it. That is useful, but it can also mean they do not question a bad premise as much as they should.

Watching this pivot-to-ai video I was struck by a thought (as it sometimes happens)… I think “we” may be setting ourselves up for the return of the internet worms, 1999-style, but…

Last month, a Brown professor wrote about discovering that a majority of his students had used AI to cheat on his take-home exam. He was irate about this “scandal,” despite the university seemingly…

AI-Assisted programming has added certain words and phrases to my technical vocabulary. Some of these are fine, most of them make me wince. They are, however, useful when talking to a clanker. All…

Rift by Brian J. N. Davis I fed some of my songs through Google's AI chatbot, Gemini, to see what would happen. That may sound like the precursor to a joke --a joke that will allow me to spend…

How the agent fleet driving my product is actually wired. One expensive planner that holds the whole graph, one cheap executor forbidden from resolving ambiguity, model tiers chosen per step against…

This week Anthropic announced that to comply with an EU AI Act they would watermark text generated by Claude models. A few tech bloggers I respect had strong negative reactions to this, and I wanted…

There’s a kind of shift occurring beneath software development right now. It has now become almost trivial to spin up a project that I wouldn’t have had time for not so long ago. I wrote…

A few months ago I started a pro subscription to Claude in an effort to learn how to use generative AI. I wanted to understand what it could and couldn’t do well. In this post I will cover a…

There’s a bizarre aspect of AI coding that I’ve been trying to put my finger on, and I think it’s this: you can basically just decide how many bugs you want your software to have…

I’m thrilled to be part of the program for Rocky Mountain Ruby 2026 in Boulder, Colorado.I’ll be speaking about…LLMs? I hope you can join us!InstiLLMent of Successful Practices in…

Google recently published a piece titled "Why Go is an ideal language for AI-assisted software engineering." I recommend giving it a read. It makes a compelling, practical point that anyone…

I recently finished the London School of Economics' programme on AI Law, Policy, and Governance. What follows is a series built from the notes I took along the way. This has been written for people…

My weekly Claude Fable limit resets every Tuesday. I usually work through it in a day or two, then I have to wait again. So I’ve noticed myself saving up more ambitious asks the rest of the week,…

We turned the MLSys'26 AE agent's core into a standalone paper reproducibility agent. Give it a paper PDF; it finds the repo, installs and runs it, and adjudicates the paper's claims.

When Accountability Goes Rogue💡I recently spoke with Dr. Zena Assaad for her podcast, Responsible Bytes. Video is above, below is an essay adapted from the conversation. "The system from…

AMD ROCm, AI inference and compute on FreeBSD via bhyve GPU passthrough, and reducing blower fan noise

If you sleep well tonight you may not have understood this lecture — Geoffrey Hinton Thought provoking lecture from the Godfather of AI Geoffrey Hinton at The Royal Institution. It pairs well with…

Anthropic announced watermarking of content across all of its new models. Google backs off the frontier, and rogue agents can hack your Gym for you.

When I started Moolah, I built it as a vibe coded, but “serious” project. I spent time planning features, thinking through design and setting up systems to make sure it was high quality.…

There are many articles and diagrams on agentic loops - I wouldn't attempt to rehash them here. What I'll start with is picking a rough form, explore mapping our SPIRIT framework to see what fits,…

This week is all about trust, and who pays for it. Two posts poke at Google’s Open Knowledge Format from opposite ends: one asks who gets to declare a piece of agent memory…

This is a recording of our AI Book Club discussion of The MANIAC by Benjamin Labatut, held August 16, 2026. This was the club's first work of fiction, and much of the discussion is about what that…

Artificial Intelligence & Society Link to heading The Future is for Everyone Zuckerberg’s positive AI future means personal agents for everyone, broad access, open source, and human…

In an older post we looked at how SynthID hides a watermark inside AI-generated images. Today, let us look at text watermarking and how it works. This week Anthropic says Claude models will use a…

Paste, CleanShot X, Wispr Flow, and 1Password do the unglamorous work that makes my AI workflow fast, visual, and safe.

Last year, when Claude Code started getting really good, I asked Claude to teach me the SLAM algorithm, and so we built a working prototype together. It wasn’t perfect, but I learned quite a lot. At…

I Don’t Code by Hand Anymore I don’t code by hand anymore, and I don’t personally know anyone else who does at this point. A few years ago…

A practical framework for reducing AI costs - auditing token spend, killing legacy models, moving workloads to open weights providers, and fixing the agent and tool inefficiencies that quietly burn…

A little over a year ago I set up a local LLM with the expectation that I would use it for all kinds of cool things. So far, it has just replaced the need for ChatGPT (chat) and Grammarly, which is a…

LLM evals in production: turning real bugs into a weekly test habit

Smolbox runs entirely in a browser tab. It is a full x86_64 virtual machine (VM) sandbox based on Alpine linux that runs under WebAssembly (WASM). It also has “AI” artificial intelligence…

Observation from heavy agent use: harness todos are fine as a scratchpad, but durable work has been more stable on external boards - especially GitHub Issues and Projects next to the code.

A coding agent built a deterministic German verb conjugator, driven by a verification loop against two morphology gold standards: from 62.8% to 99.2% agreement.

I recently watched Lauren Celenza’s talk, Living Through an AI Takeover Without Losing Your Soul from Beyond Tellerrand conference. The title hooked me, mainly because the soul is precisely the thing…

This journey starts like most of the things I've been doing this year, as an experiment to run a website 100% unattended with an agent. Ideally this post won't go down as an epitaph in a future where…

You don't need a new sandbox product to keep AI agents out of your secrets. Bubblewrap is a decade old and already does this.

Writing this from Orania Hotel in Berlin, sipping a negroni and doing a little people-watching as people ride, walk, scoot, and drive past the floor to ceiling windows. It’s a nice hotel,…