A New Kind of VCS: Parallel Branch Checkouts for Humans and AI
Version control systems (VCS) have been around for decades, and Git has dominated for over fifteen years. But as development teams begin to incorporate **AI coding assistants** into their workflows, we’re running into a fundamental limitation in Git’s design: only one branch can be checked out in a working directory at a time.
A New Kind of VCS: Parallel Branch Checkouts for Humans and AI
Version control systems (VCS) have been around for decades, and Git has dominated for over fifteen years. But as development teams begin to incorporate AI coding assistants into their workflows, we’re running into a fundamental limitation in Git’s design: only one branch can be checked out in a working directory at a time.
This constraint may seem minor, but it introduces friction when you want an LLM working on one branch while a human developer works on another, both operating on the same repository simultaneously. Today, the workaround usually looks like this:
- Clone the same repository twice on disk.
- Checkout branch
Ain clone #1 for the LLM. - Checkout branch
Bin clone #2 for yourself. - Sync them by pushing to a remote (or fiddling with local remotes between the two clones).
This works, but it’s clunky. It means doubling disk space, increasing setup overhead, and often requiring network pushes just to let two branches “talk” locally. For workflows that need fast back-and-forth iteration between humans and AI, Git feels a little… dated.
Rethinking the Workspace Model
Imagine a new VCS—let’s call it Branchspace—designed with the assumption that multiple working branches can coexist in the same repo checkout.
Instead of tying a working directory to a single branch, Branchspace introduces the concept of parallel worktrees as first-class citizens:
- You can open
mainandfeature-ai-autogenin parallel. - Each branch has its own isolated working tree, but they share the same
.branchspaceobject database under the hood. - No need for multiple clones, symlinks, or complex Git worktree setups.
- Both human and AI agents can make changes side by side, locally, without stepping on each other.
The repo itself becomes a kind of multi-branch workspace, where switching context is as simple as changing your working directory path (e.g., repo/.branches/main vs repo/.branches/feature-ai-autogen).
Why Does This Matter?
-
If you want an LLM to generate scaffolding or refactor code in , you can let it work uninterrupted while you polish code in . You don’t have to babysit branch checkouts or pause your flow while the AI works.
