Headcode began as a way to learn UK rail data and how to use an LLM properly on something real. The product is a maybe; the research-to-PRD-to-tasks workflow I built alongside it, with review at every layer, is now how I work.

Introducing jjw, a small Go CLI for managing jj workspaces with bookmarks and lifecycle hooks, built to make parallel LLM coding agent workflows easier.

After losing a PostgreSQL database by blindly following AI-generated commands, I developed a structured approach to working with AI that separates generation from judgement