Four AI agents fix eleven app bugs in parallel, with one merge conflict

A developer ran four Claude Code agents at once to work through eleven bug reports in their app, with the lead session first confirming which reports still applied to current code. Bugs were grouped so that no two groups touched the same files, leaving a single overlapping file that had to be merged by hand. Each agent worked in its own git worktree built from the latest main branch, and on Windows the worktrees shared the main install's node_modules through a junction. The agents wrote code and ran typechecks without committing, while a lead session ran the full test suite and reviewed every diff in one release worktree. That review caught several mistakes, including a test storing a matcher instead of a value and a test expecting removed behaviour, before anything was committed.