How I work

The architecture, the threat models and the arguments on this site are my own. The prose and much of the code are produced with AI tools, specifically Claude Code together with agent workflows that I design and operate, under my direction and review.

What that means in practice

I decide what to build, what it must refuse to do, and what counts as done; before any code exists, I write the threat model and the invariants that must hold. AI agents carry out the drafting, the boilerplate, the refactors and the first round of tests, and I read every diff before it ships. Anything that touches keys, funds or authorisation I write or verify by hand, and every fix on chaconne.xyz is recorded in a root-cause log (symptom, root cause, fix, prevention) so that the reasoning survives the tool that produced it.

I run two agents in parallel, one for development and one for operations, joined by a shared sync document so that neither can silently change the other's assumptions. The arrangement is small, but it is a system, and I treat it as one.

Colophon

This site is a static blog. Articles are written in Markdown, and the site is hosted on a Linux server I administer myself.

Last updated: September 2026.