← Back to feed

Mitchell Hashimoto joined GitHub in February 2008. He was user number 1,299. He has visited GitHub almost every single day for over 18 years. He is leaving.

His project Ghostty is moving off GitHub after repeated infrastructure failures turned everyday maintenance into blocked reviews, stuck merges, and failed automation. The proximate cause is not a single outage. It is a structural change in what software development platforms are now asked to absorb.

The Load Shape Has Changed

AI coding agents do not work the way human developers work. A developer who once pushed a few careful changes per day can now push many AI-assisted iterations in the same span. Every iteration wakes CI pipelines. Every iteration triggers indexing, storage writes, review system events, and retry logic. The event volume per developer has compounded.

GitHub was architected for human pacing. The platform’s core infrastructure — its CI execution layer, its pull request review queues, its search indexing, its storage backend — was built around the assumption that code arrives at the speed humans can write and review it. That assumption no longer holds.

The bottleneck is no longer writing code. It is absorbing code.

What This Means for Developer Platforms

Hashimoto’s departure is a signal worth tracking. He is not a casual user frustrated by a bad week. He is one of GitHub’s oldest power users, the author of tools that tens of thousands of developers rely on, and someone who has stayed through every prior platform rough patch. The decision to move Ghostty reflects a considered judgment that GitHub’s infrastructure failures are structural, not incidental.

The immediate consequence for Hashimoto is losing integrations, automation, and community expectations built around GitHub-specific workflows. The cost of moving is real. The fact that he is paying it anyway is the data point.

The Wider Pattern

This is not an isolated complaint. The pattern — AI coding agents generating repository events far beyond human-authored baselines — is observable across every platform that hosts software projects at scale. Pull request counts are up. CI job counts are up. Review queue depths are up. Merge conflict rates are up.

GitHub has not publicly commented on its infrastructure capacity relative to AI-generated load. What is clear from operator-level reports is that the platform’s throughput ceiling is being tested in ways it was not two years ago.

Developer infrastructure is the next AI capacity problem after compute.