GitHub Hits 275 Million Commits Per Week. Infrastructure Is Cracking.

275 million weekly commits. Projected 14 billion for 2026. GitHub has shifted to an "Availability First" posture, which is corporate speak for "we're running out of headroom."

The platform built for human-speed collaboration is being stress-tested by machine-speed development. AI agents don't take lunch breaks. They don't context-switch. They commit code as fast as they can write it, and with tools like Codex's remote-control mode making headless agent sessions trivial, the volume is only going up.

GitHub has been dealing with escalating outages as autonomous agents hammer APIs and create merge conflicts at machine speed. Rate limiting is getting tighter. The fundamental assumption that version control could absorb development velocity at any scale is being tested in real time.

The Register reported that Git's sequential stop/go workflow model wasn't designed for this. The 180 million users across 630 million repositories create infrastructure pressure that concentrated AI agent activity amplifies. Proposed solutions range from local-first Git to global mirroring to agent-native version control tools like GitButler.

Here's what worries me more than the infrastructure. When agents commit faster than humans review, what does code quality mean? A separate Register piece called AI-generated code "pain waiting to happen," documenting how manager enthusiasm for AI coding tools has outpaced developers' ability to learn them. The maintenance time bomb is real. Code entering production faster than teams can review it will surface as incidents, not metrics.

If you're running automated agent workflows on GitHub, watch your rate limits. Consider whether your review process can actually keep pace with agent-speed commits. And start thinking about what version control looks like when the majority of commits aren't written by humans. We're not there yet. But 275 million a week says we're closer than most people think.