Why should ai need this? A linear backlog is enough, a cache, for everything else they can create it new in a short time.
Other commenters mentioned worktrees, which let you check out different branches at the same time from a single local repo. That's convenient, but not required.
Git always supported "fast cloning" local repos as well. You just "git clone" from one directory to another. Then they are independent and you're free to decide what to merge back.
These days, agents can also fork their containers or VMs as often as required too, with copy-on-write for speed.
So that's four ways to work on multiple branches in parallel using Git that we already use.