We Had to Write Docs for AI: llms.txt Changed Everything

Most developers write documentation for humans. While building my JavaScript framework, I ran into a problem I didn't expect: the framework worked but AI couldn't use it. Not "wasn't perfect." Not "made small mistakes." It completely failed to build even basic apps correctly unless it had the source code of the framework available. The Moment Things Broke After years of work, I finally had a stable system: a custom scanner, parser, interpreter, a template engine with components, a signal-based reactivity system, and around 600 tests covering edge cases. I thought I was done. So I tried something simple: "Build a todo app using this framework." What I got back looked confident, but was completely wrong. Wrong syntax. Wrong mental model. Invented features that didn't exist. This wasn't a bug in the framework. It was a documentation failure. README Is Not Enough Anymore Traditional documentation is designed for humans: narrative explanations, gradual onboarding, examples mixed with storyt