DeckSpaceRunning several AI coding agents at once.
Three guides about the part nobody writes down: what actually happens when more than one agent is working in your repo, and what has to be true before that is safe.
They state limits as well as capabilities. Where a setting does nothing on a given CLI, the page for it says so.
guide
Run multiple Claude Code instances at once
A second terminal works. What it does not give you is any view of what the other session changed — and both of them stage into the same git index.
guideSeveral agents on one repo, without collisions
The git-worktree-per-agent pattern, by hand and then automated. Where the roster cap is really enforced, and what the Sandbox setting does not cover.
guideWhat breaks when agents live in terminal tabs
Two tabs are fine. Around four you stop knowing which agent is waiting on you, what each one already changed, and whether anything was reviewed.