Git tracks files. Not context.
That's the problem.
When an AI agent writes code, Git sees the diff. It doesn't see which model wrote it. It doesn't see the prompt or the temperature setting.
If a bug appears, you need to know which agent introduced it. Was it Claude Code? Gemini?
Git gives you a hash. Not an answer.
Git Was Built for Humans
Git was built for Linux kernel development in 2005. Now it has to handle AI agents writing half your codebase.
They generate, iterate, and sometimes break things in ways humans wouldn't.
Version control is becoming a coordination layer — not just tracking changes, but orchestrating humans and agents.
That's a fundamentally different job.
Where It's Already Breaking Down
Stacked branches are painful. Most teams work on multiple features simultaneously.
Git forces you to choose one branch. AI agents need parallel work. Git was designed for sequential patches.
The plumbing might need an upgrade.
But Do We Actually Need Something New?
Git works. It's everywhere. Every CI pipeline is built around it. Every developer knows it.
We can build better interfaces on top — GitHub, GitLab, Graphite already do.
But maybe the pipes themselves are too narrow.
The $17 Million Question
A company just raised that much to build what comes after Git. The pitch: Git wasn't built for this era.
I'm not convinced yet.
But a16z doesn't throw $17M at small problems. They see a shift.
If they're right, we're not just talking about a new tool. We're talking about rebuilding how software gets built.
Is Git enough for the AI era, or do we need to rebuild version control from scratch? 👇
Top comments (0)