Standard Git allows only one working directory per clone. If you need to work on two branches simultaneously — say, a feature branch and a hotfix — you have to either stash changes, use multiple clones, or switch back and forth. Git worktrees solve this by creating additional working directories that share the same .git object store and history while having independent branch checkouts and working trees.
For AI agent workflows, worktrees are valuable because each agent can work in its own directory without interfering with other agents or the developer's current working state. Agent A checks out feature/payments in /repo-worktrees/feature-payments; agent B checks out feature/auth in /repo-worktrees/feature-auth. Both agents run simultaneously without interfering with each other or with the developer's main checkout.
Worktrees share the object database, so creating one is nearly instantaneous and takes no additional disk space for the history — only the working files. This makes them practical to create and destroy per-task rather than per-project.