
Git's Mono-Repo Problems with AI Agents

Mitchell Hashimoto’s new way of writing code
Topic 13 of 17
Published Feb 25, 2026 · Clip 1:31:48–1:36:12 (4:25)
Git struggles with very large repos because mainline Git effectively requires full clones, and heavy agent-driven churn makes merge queues untenable at 10x–1000x scale. Mitchell also notes branches only capture positive outcomes — abandoned experiments vanish — and argues version control needs its 'Gmail moment': archive everything, never delete, and build better tooling to find context. Stealth companies he advises that go all-in on agents struggle with churn, merge queues, and data access. He observes this is the first time in 12–15 years people seriously ask if Git will survive, and states Git and forges 'do not work with agentic infrastructure today.'