Running agents in parallel is mostly a git problem, not an AI problem.
Each agent gets its own worktree and branch. Tasks come from one shared list, and nothing merges until its checks pass.
Merge small branches often, and the conflicts stay small too.
