Colophon
How the posts are written
Every post here is a record of real work: something I built or broke, what I tried, and what the numbers said. I do that work with AI coding agents, and the posts say so where it matters — what I asked the agent, what it came back with, where it was wrong and how I caught it.
The posts are also drafted with an agent, from the session the work happened in. The agent gathers what happened; I decide what the post is about, what stays in, and whether it is worth publishing at all. Every number is one I measured, every claim is one I would make out loud, and I read the final draft as a stranger before it goes up. If something is wrong, that is on me, and I would like to know.
The illustrations are generated from a scene I describe for each post, in one house style, with the same recurring character. He always wears a hat. That is a rule, not an accident.
How the site is built
Posts are Markdown files in a Git repository. A push to main builds a Next.js site into a container and deploys it to Google Cloud Run, which scales to zero when nobody is reading. Every page is rendered at build time; there is no database, no CMS, and nothing loads on the client that does not have to. Code is highlighted at build time too, so no JavaScript ships for it.
The site is open on GitHub, including the scripts that scaffold a post, generate its banner, and check it for things that should not leave the machine.
What is collected
Google Analytics, for page views and referrers, so I can tell what people read. Nothing else: no comments, no accounts, no tracking across sites. The RSS feed carries full posts, so a feed reader never has to visit.