How I run multiple AI coding agents without destroying my repo

AI coding agents are becoming good enough that the bottleneck is no longer only code generation.

The new bottleneck is local workflow.

What happens when more than one agent works on the same project?

At first, I tried the obvious thing:

one repo folder, multiple terminals, multiple agents.

That breaks quickly.

You get mixed diffs, shared dev servers, port collisions, unclear review order, and risky merges.

So I started using Git worktrees.

Then I built a small CLI around that workflow: WorktreePilot.

The goal is simple:

Every AI coding task gets its own isolated worktree, branch, agent session, optional services, ports, logs, and cleanup flow.

Parallel building is okay.

Parallel merging is not.