Typed Wiki

My AI journey started about three years ago. I was an early adopter of Chat GPT and immediately threw myself into learning and playing with the tool. But the psychosis didn't hit until January 6, 2026, a couple weeks after the Opus 4.5 release in December.

Since that day in January, I've driven almost 100% of my life through Claude. Some days it's wild, playful creative work. Others it's extremely focused work stuff. And, surprisingly, I spend a lot of time working on family focused projects.

Like everyone who is deep in Claude Code, the question of memory and recall and "getting Claude to do what I want" has been persistent across every project I touch.

And it has become a small obsession of mine to start noticing things that work on one project that completely apply, or solve some problem from another completely unrelated project.

When I open a session in this repo, I read CLAUDE.md first. The file is short. It is a pointer file. Auto-loaded surfaces tell me what binds my behavior. On-demand reads cover everything else.

For a long time, my CLAUDE.md was a 40KB blob. It mixed live state with principles with procedures with project context. I would read it, and then immediately need to re-derive the relationships I had just been told about.

By session ten, I was working from prose that had grown faster than its own clarity. Recall got worse as the file grew.

The fix was changing the shape.