Isolation (via Worktree)

What is Isolation (via Worktree)

Isolation via worktree is the practice of giving each agent its own git worktree specifically so that its file changes cannot collide with another agent working on the same repository at the same time. The worktree feature itself is the mechanism, isolation is the reason it gets used in a multi-agent setup.

This becomes necessary the moment more than one agent is touching the same codebase concurrently. Without separate directories, two agents editing overlapping files risk one overwriting the other's uncommitted work, or two agents reading a half-finished edit and reasoning from an inconsistent state. Separate worktrees remove that risk at the filesystem level rather than relying on agents coordinating perfectly around each other.

This pattern is the default assumption behind parallel agents and fan-out: dispatching several agents at once onto one repository without isolating them is treated as a mistake, not an optimization. Once an isolated agent finishes, its branch merges back into the shared history, and the worktree is cleaned up, the isolation was temporary scaffolding for the duration of the work, not a permanent split of the codebase.