Building Git infrastructure for agent-scale development

This explains why Git infrastructure must be rebuilt to support "agent-scale" development, where humans and autonomous agents push millions of changes concurrently. It reports dramatic growth in Git activity: monthly events doubled from 218.2 billion to 473.3 billion (Sept 2025-Aug 2026), pushes rose 4.9× (0.69B to 3.35B/month), and a single month saw 7.38 billion commits - over five times year‑over‑year. The busiest repository handled roughly a billion requests in one month. Those numbers motivate an architecture overhaul that must happen while the platform stays live.

The technical argument centers on three concrete scaling problems: per‑agent commit turnaround (agents checkpoint after nearly every action, so single‑push latency limits their speed), vastly higher sustained write throughput as thousands of agents create many branches, and merge contention on a single reference as trunk‑based workflows and merge queues funnel work onto one ref. The solution focus is on building infrastructure that supports sustained, concurrent reads and writes with low single‑push latency, and on principled design changes to remove single‑point bottlenecks so large teams and fleets of agents can operate at their busiest.