Multi-Repo AI Coding: Move the Context to Linear
A while back I wrote about using requirements markdown files as context for AI coding sessions, and then about why loading that context fresh at the top of a session works so well. That workflow is still the foundation of how I code with AI. But recently I hit its ceiling.
The ceiling is called "a project that spans multiple repos." An API repo, a frontend repo, a firmware repo, an infra repo, and multiple Claude Code sessions working across them, sometimes at the same time. The requirements-file-in-the-repo approach starts to crack here, because the question becomes: which repo? So I moved the project-level context out of the repos entirely and into Linear. It has been one of the best workflow changes I've made this year.