GitHub shifts to 30x scale as agentic coding starts breaking ops

GitHub's latest availability update reads less like a standard status post and more like an infrastructure warning from the front line of the AI coding boom. In the April 28 post, the company says it began a 10x capacity program in October 2025, then realized by February 2026 that it had to design for 30x today's scale. The reason is not abstract traffic growth. GitHub says agentic development workflows accelerated sharply from the second half of December 2025, changing how often repositories, pull requests, automation, and APIs are being hit.

That matters because a modern pull request is not one isolated event. GitHub lays out the chain plainly: Git storage, mergeability checks, branch protection, Actions, search, notifications, permissions, webhooks, APIs, background jobs, caches, and databases can all get touched by a single workflow. When usage rises everywhere at once, small inefficiencies stop being small. Queues deepen, cache misses turn into database load, and a feature that looked independent starts amplifying pressure somewhere else.

The company used two recent incidents to show what that looks like in practice. On April 23, a merge queue regression caused squash merges in multi-PR merge groups to produce incorrect merge commits, affecting 230 repositories and 2,092 pull requests. On April 27, an incident tied to GitHub's Elasticsearch subsystem disrupted several search-backed experiences across pull requests, issues, and projects. GitHub says there was no data loss and Git operations and APIs were not impacted, but the failure was still broad enough to break important parts of the user interface.

The larger signal is straightforward: AI-assisted software development is now stressing the platforms beneath it, not just the models on top. Reliability, isolation, and blast-radius reduction are becoming product features in their own right. If agentic coding keeps compounding traffic across repositories, search, automation, and review flows, the winners in developer tooling may be the vendors that can make all of that scale without turning every growth spike into an incident report.