In April, Mitchell Hashimoto pulled Ghostty off GitHub after 18 years on the platform, with a blunt verdict: it’s “no longer a place for serious work.” He isn’t a crank, and Ghostty isn’t a fringe repo. When someone like that leaves after 18 years, it’s not a complaint. It’s a leading indicator.
This isn’t about agents flooding the pull-request queue. I’ve written about that throughput problem and how it hit release engineering already. This is the quieter, more structural story underneath it: the platform itself is being neglected by the company that owns it.
The Neglect, Quantified
You can argue about vibes. The operational record is harder to wave away.
- No CEO since August 2025. Thomas Dohmke stepped down and was never replaced. GitHub’s leadership now reports directly into Microsoft’s Core AI team.
- 48 major outages in 12 months. More than 330 hours of downtime, with incident counts up well over 50% year on year. Actions, Copilot, pull requests, core Git operations, all of it.
- An Azure migration that capped capacity right as agent traffic exploded, turning a scaling problem into a reliability one.
A platform without an owner-level champion, run as a line item under a different org’s priorities, degrades exactly like this. Not in a dramatic collapse. In a thousand 20-minute outages nobody is accountable for.
The Conflict of Interest
Here’s the part that makes the neglect rational rather than merely sad. GitHub’s owner sells Copilot. Platform reliability and Copilot revenue are not the same goal, and when they compete, you can see which one wins.
GitHub is no longer being run as the world’s code host. It’s being run as a feeder and a surface for an AI product, and a product whose pricing keeps lurching while it has quietly fallen behind Cursor in the market it was supposed to own. The host is being neglected to subsidise a horse that isn’t even winning.
The Exodus
Hashimoto is the loud example. The quiet ones matter more:
- Zig, cURL, and Godot are reducing reliance, mirroring, or moving infrastructure off GitHub.
- The pattern is maintainers of load-bearing projects acting after outages, not before, which means the trust erosion is already priced in.
Migration is painful and network effects are real, which is exactly why these moves are signal. People don’t leave 18 years of issues, stars, and muscle memory over a bad week. They leave when they’ve concluded the bad weeks aren’t going to stop.
— Mitchell Hashimoto, on moving Ghostty off GitHubIt is no longer a place for serious work.
The Stewardship Bet
The deeper issue is that we made a bet most of us never consciously placed. Centralising nearly all of open source onto one platform was a wager that the steward would stay aligned with the commons it hosted. For years, Microsoft was a genuinely good steward. The acquisition fears didn’t materialise. People relaxed.
But stewardship is a standing choice, not a one-time grant. The incentive that mattered, “host the world’s code well,” has been quietly replaced by “convert the world’s code into AI product.” Those overlap less every quarter. The commons didn’t move. The steward did.
What This Doesn’t Solve
The honest counterweight, because “GitHub is dying” is the kind of headline that ages badly.
- It won’t die quickly. The network effects are enormous. “Too big to leave” is real, and most projects will stay out of pure inertia.
- The alternatives have costs too. GitLab, Codeberg, sourcehut, and self-hosting each trade one set of problems for another. Distributed git was supposed to make the host disposable; in practice, the issues, CI, and social graph aren’t.
- Agent traffic would strain anyone. Some of the reliability pain is genuinely the commit-volume surge hitting infrastructure built for a slower era. That’s not unique to GitHub.
- Microsoft could still re-engage. Appoint a real CEO, fund reliability, treat the platform as the asset it is. The bet isn’t lost. It’s being called.
The lesson isn’t “leave GitHub.” It’s that depending on a single steward for shared infrastructure is a position you have to keep re-underwriting. The time to know your exit is mirrored and your CI is portable is before the steward’s incentives drift, not after the third outage of the week.
Closing
Review broke under the flood. Release broke after it. Now the stewardship layer is the one buckling, and it’s the one no feature flag can fix. GitHub spent a decade earning the right to host nearly everything. It can lose that the way trust always goes: slowly, then all at once, while the owner is busy looking at something else.
Related Posts
