Most of what I do with AI is not prompting. It is the layer around the models: what runs, against which repo, under what written goal, and how anyone knows afterwards whether the work happened. That layer runs many unattended coding sessions in parallel across about seventy tracked pieces of work, with state in files…
The detail I can't stop turning over isn't the zero-days. It's May 8: an agent gets stuck because a file was…
The number that stopped me wasn't 89.0% on ARC-AGI-1. It was $0.02 per task. I've spent the last year writing…
I read "Proxmox for Arm64" and immediately thought about the Pi bolted behind my nav station. Then I read the…
keyv is not some exotic dependency. It's the thing three layers under the thing you actually installed —…
I click approve on agent commands all day. Turns out I'm bad at it. In a browser game where you play…
Yesterday was a phone day. Most of the work went into cityheatmap, which is now a live site at cityheatmap.com. The morning was about coverage: measuring how many readable opening hours each city actually has before it gets offered in the selector. The…
read the day →- TIL Archive the inputs to a reduction, not its result
- News The bottleneck is me, not the model
- Decision Why I am not founding a Nordic depth database
- News Celld puts Durable Objects on my own hardware, and the bucket is the whole control plane
- Finding The missing-receipt report that doesn't exist
- News The Off Switch That Wasn't
- Diary The day started with a question I couldn't answer: why was Clemens Bar…
- News A moderation model I can actually host on the boat
- News Kimi K3 is close enough that I'll rerun my agent evals
- Retro What the fleet actually shipped in July
- Diary Yesterday was a big day across a lot of surfaces, and most of it was…
- News Cheap tokens only matter if the model shuts up