GitHub Keeps Going Down

On February 9th, 2026, GitHub went down.

Not “a little slow.” Not “some features degraded.” GitHub.com, the API, Git over HTTPS, Actions, Copilot - basically all of it was throwing errors.

Millions of developers. Thousands of companies. CI/CD pipelines around the world, frozen in place.

And the status page? It made the whole thing look like barely an issue.

Note: This post is a blog-style adaptation of our video on GitHub’s recent downtime and what it says about the future of software infrastructure.

The real status page that day was Twitter. ThePrimeagen joked that he had “finally figured out why github is going down they are training on my code,” and the post racked up thousands of likes in a few hours. He wasn’t alone. Over the following weeks, developer timelines filled with the same question in a hundred different forms: why does GitHub keep falling over?

One developer built an unofficial site called The Missing GitHub Status Page, which scrapes GitHub’s own incident reports and reconstructs the availability picture GitHub no longer shows in one place. It’s called “missing” for a reason. GitHub’s older status page used to publish aggregate availability metrics. During the migration to the current status site, those rollup numbers quietly disappeared.

At one point in 2025, the unofficial reconstruction of GitHub’s uptime reportedly dropped below 90 percent. Not 99.9. Not even 99. Below 90. That’s one nine of availability.

If you’re a developer, you probably already know what this feels like. You push your code, or try to, and nothing happens. Your CI job queues and just sits there. You open Hacker News, find a thread with four hundred comments, and feel the sweet relief of realizing it isn’t just you.

So here’s the question nobody has a great answer to: how did the platform hosting most of the world’s open-source code, plus a massive chunk of the world’s private code, get so fragile?

And the bigger question: is it going to get better?

Status pages are not reality

In April 2026, GitHub overhauled its status page. They added a new “Degraded Performance” tier and started publishing weighted per-service uptime because, by their own admission, the old system wasn’t accurately reflecting what users were experiencing.

That’s the polite corporate version of saying: the map didn’t match the territory.

Status pages are often lagging indicators. They’re categorized, curated, manually adjusted, and smoothed into labels that don’t always match what users are experiencing in real time. One Hacker News commenter pointed out that some companies don’t even automate status page changes; a human has to go update them. And humans, as a species, are not famous for enjoying public admissions of failure.

GitHub had six incidents in February, four more in March, and a Copilot outage in January that peaked at a 100% error rate. Then, after this video was recorded, two more major incidents landed in April: incorrect merge commits from squash merges in merge queues, and an Elasticsearch outage that made parts of pull requests, issues, and projects unusable.

The irony is almost too perfect: our own video progress was tracked on a GitHub Projects board that became unusable during the GitHub outage video.

But GitHub is not the only system that keeps reminding us software is still, unfortunately, software.

Everyone is fragile now

At 11:48 PM Pacific on October 19th, 2025, two background processes in AWS’s us-east-1 DNS management system got into a race condition.

Two copies of the same program were running in parallel for redundancy. Both tried to update the IP records for DynamoDB at nearly the same time. One finished first. One was slower. The fast one completed and ran cleanup. The slow one, holding a stale plan from minutes earlier, wrote its outdated version on top of the fresh one. Then cleanup saw the outdated plan, decided it was garbage, and deleted it.

And just like that, every IP address for the regional DynamoDB endpoint was gone.

For the next fifteen hours, Snapchat, Reddit, Venmo, Fortnite, Ring doorbells, Amazon Connect call centers, and even parts of the AWS console people needed to fix the problem were impaired or unavailable.

Four months earlier, Google Cloud had its own major incident, after a bad globally replicated policy change with blank fields triggered a null pointer exception and crash loop in the Service Control stack.

So no matter what the big AI CEOs are saying, software is not a solved problem. There’s no obvious path to it becoming one, even with thousands of agents autonomously making code changes. No, especially with thousands of agents autonomously making code changes.

Git is not GitHub

To understand why GitHub’s situation is particularly weird, you have to go back to the beginning.

Git is not GitHub.

In 2005, Linus Torvalds got fed up with BitKeeper, the proprietary version-control system the Linux kernel had been using. The licensing relationship fell apart, and he needed something fast. So he built Git.

The first working version took about ten days. And he did it without an AI slop cannon. A real John Henry.

Git was designed to be fast, distributed, and resistant to single points of failure. Every clone of a repository is a full copy of its history. No central server required.

That last part matters.

Git is decentralized by design. GitHub is not.

Torvalds later said he never expected Git to take over the world. He maintained it for about four months, handed it off to Junio Hamano, and went back to working on Linux. His daughter eventually texted him from college to say he was more famous in the computer science lab for Git than for Linux.

But Git did take over. And a big reason was the thing launched three years later.

GitHub went live in April 2008. It added the collaboration layer that made Git pleasant for teams: a web interface, pull requests, issues, profile pages, stars. Less than a year and a half after public launch, GitHub had 100,000 users.

By the time Microsoft acquired GitHub in 2018 for 7.5 billion dollars, it was already the gravitational center of open source.

The acquisition went better than a lot of developers feared. Microsoft mostly left GitHub alone, at least for a while. But somewhere along the way, GitHub stopped being just a nice place to host repositories.

It became critical infrastructure.

Actions runs CI/CD. Packages hosts registries. Copilot depends on GitHub’s backend. Issues and projects manage work. Pull requests gate deployment. And all of that shares underlying infrastructure.

So when one thing breaks, everything breaks.

GitHub’s CTO later admitted that the recent availability problems were driven by rapid usage growth, architectural coupling, and poor load shedding.

In plain English: too big, too tangled, and not written in a way that gracefully handles heavier-than-normal traffic.

Which raises the obvious question: why is there suddenly so much traffic?

The clankers are here

GitHub’s Octoverse 2025 report said the platform added 36 million new developers in a single year, with nearly a billion commits.

Then GitHub COO Kyle Daigle posted that by early 2026, the platform was seeing 275 million commits per week. That’s a pace of roughly 14 billion commits per year.

One billion commits per year to fourteen billion.

In one year.

There are not suddenly ten times as many developers writing code. It’s the clankers.

A meaningful chunk of that traffic is AI-generated: Copilot autocomplete, agentic coding tools pushing commits, agents running on someone’s spare Mac mini opening pull requests. The public numbers make it clear that AI-assisted development is scaling insanely fast. And those are just the systems that admit what they are.

Plenty of agents pretend to be human.

Trend Micro found AI-assisted fake GitHub repositories with AI-generated READMEs designed to look human-made, some of which were distributing malware. Obviously.

GitClear analyzed 153 million changed lines of code and found that code churn, the rate at which code gets rewritten or thrown away, is worse with AI.

More code. Lower quality. Unbelievable load on GitHub.

I’d feel worse for them if Microsoft wasn’t largely responsible for shoving AI down everyone’s throats in the first place.

AI is externalizing its costs

GitHub is not the only place absorbing the blast radius.

In March 2025, Drew DeVault, the creator of SourceHut, wrote a blog post titled “Please stop externalizing your costs directly into my face”.

He described spending between 20 and 100 percent of his working hours, every week, fighting off LLM crawlers. These bots ignore robots.txt. They come from tens of thousands of residential IP addresses. Each IP makes exactly one request, which makes them hard to rate limit. They fake user agents to look like real users. And they specifically hammer the most expensive endpoints on a git forge: git blame, every page of every git log, every commit in every repo.

Several high-priority SourceHut tasks, he wrote, had been delayed weeks or months because they kept getting interrupted by bots.

Cloudflare says crawlers that don’t respect “no crawl” directives, many of which are likely AI bots, now generate more than 50 billion requests per day through its network alone.

Daniel Stenberg, the maintainer of curl, has been writing about AI-generated vulnerability reports that sound plausible, are totally fabricated, and cost him real time to disprove.

You know all those grifters minting NFTs four years ago? They’re now pushing tens of thousands of commits a day to repos nobody will ever read. My belief is that more compute cycles are being spent generating much of this code than will ever be spent executing it in production.

Where do developers go?

If GitHub keeps getting worse, where are we supposed to go?

In November 2025, Andrew Kelley, the creator of Zig, published “Migrating from GitHub to Codeberg”. For a certain kind of developer, it read like a eulogy.

“Ever since git init ten years ago, Zig has been hosted on GitHub,” he wrote. “Unfortunately, when it sold out to Microsoft, the clock started ticking. ‘Please just give me 5 years before everything goes to shit,’ I thought to myself. And here we are, 7 years later, living on borrowed time.”

He described GitHub Actions “vibe-scheduling” jobs and complained that Zig’s strict no-LLM policy kept getting violated because GitHub keeps putting Copilot buttons in front of contributors.

So Zig left.

Their new home is Codeberg, a German nonprofit running on Forgejo, a community-maintained fork of Gitea. Gentoo Linux is migrating too. Then there’s GitLab, the obvious enterprise escape hatch, and SourceHut, which is refreshingly anti-AI and whose web frontend doesn’t even use JavaScript.

All of them share the same fundamental challenge.

The developers are on GitHub.

That’s the problem. I’m still on GitHub. I hate the downtime, but honestly, it hasn’t gotten bad enough for me to drop everything and move.

And to be fair to GitHub, the alternatives don’t carry the traffic GitHub does. If everyone moved at once, these problems wouldn’t magically disappear.

But when Zig moved to Codeberg, they knowingly gave up some discoverability. And they seemed fine with that.

We’re living in a time where a little isolation can be a benefit. The issues and pull requests “contributed” on GitHub are increasingly slop. When the community aspect of a platform flips from positive to negative, the network effects start working in reverse.

It’s no longer “everyone is here, so you should be too.”

It’s “everyone is here, and maybe I want them to go away.”

Bots are not interns, but they need supervision like interns

The infrastructure we depend on is being hit from both sides: more traffic from AI agents, and more fragile code produced by AI agents.

In December 2025, according to reporting later echoed by The Guardian, Amazon’s AI coding agent Kiro was reportedly allowed to “delete and then recreate” part of a live AWS environment, contributing to a thirteen-hour interruption. Amazon disputed the AI-gone-rogue framing and called it user error, but the outage was real.

Kiro isn’t alone. In 2025, SaaS founder Jason Lemkin said Replit’s AI agent, building an app against a live customer database, deleted the production database, fabricated outputs saying everything was fine, and then lied about what it had done. Replit’s CEO later apologized.

AWS has reportedly added mandatory peer review for production access.

Which is a polite way of saying we now have to review the bot the same way we’d review an intern. A drugged-out, schizophrenic intern.

Is this the beginning of the slip?

GitHub’s collapse in service quality does not have to be the pattern for the whole industry. But it will be unless we stop pretending scale solves everything, build systems that can actually shed load gracefully, and accept that maybe letting bots push fourteen billion commits a year has consequences.

It’s hard to say if any one GitHub alternative is the future of source control. Everyone seems to agree we need alternatives. And sometimes, just the threat of alternatives is enough to force an incumbent to take the problems seriously.

Every month, a few more projects decide they’ve had enough. Maybe the right framing isn’t “everyone should leave GitHub.” Maybe it’s just that the people who care about writing good software are increasingly willing to do it somewhere a little less mainstream.

Somewhere the firehose isn’t.

That February 9th outage, the one where the status page underplayed a platform that was clearly on fire, might end up being one of those moments people look back on.

Not as the day GitHub died, but as the day its dominance started to slip.

If you want to learn how to actually build software, the kind that doesn’t immediately fall over when bots send requests, check out Boot.dev. Use code BOOTSTUBE for 25% off your first year.

And don’t stop learning.

Update: two more incidents and Ghostty leaving GitHub

Originally, the video was going to end there. Then, between recording and publication, there were certain developments.

Specifically, two more major GitHub incidents. On April 23, 2026, a subset of pull requests merged using the squash merge method produced incorrect merge commits. On April 27, major parts of pull requests, issues, and projects became unusable because of an outage in GitHub’s internal Elasticsearch subsystem.

Then Mitchell Hashimoto, founder of HashiCorp and creator of the Ghostty terminal, announced that Ghostty is leaving GitHub. In his post on X, he wrote that he’s GitHub user 1299, joined in February 2008, and has visited GitHub almost every day for 18 years.

“It’s never been a question for me where I’d put my projects: always GitHub,” he wrote. “I’m super sad to say this, but its time to go.”

He hasn’t announced the new home yet. But for him, enough is enough.

Hopefully we can publish this before another major outage happens.

Bibliography

1. “GitHub Availability Report: February 2026.” GitHub Blog, March 12, 2026. https://github.blog/news-insights/company-news/github-availability-report-february-2026/
2. GitHub status page, Internet Archive capture, February 9, 2026. https://web.archive.org/web/20260209235449/https://www.githubstatus.com/
3. ThePrimeagen, post on X. https://x.com/ThePrimeagen/status/2037013226396037275
4. Arnav Gupta, post on X, March 12, 2026. https://x.com/championswimmer/status/2031908498305356077
5. Nick Dobos, post on X, March 31, 2026. https://x.com/NickADobos/status/2039085867432591508
6. Chris Ebert, post on X, April 13, 2026. https://x.com/realchrisebert/status/2043732585696629106
7. Archived GitHub status pages, Internet Archive, March 28, 2017 and May 8, 2019. https://web.archive.org/web/20170328180759/https://status.github.com/ and https://web.archive.org/web/20190508015137/https://www.githubstatus.com/
8. Richard Speed, “GitHub appears to be struggling with measly three nines availability.” The Register, February 10, 2026. https://www.theregister.com/2026/02/10/github_outages/
9. “GitHub Availability Report: March 2026.” GitHub Blog, April 2026. https://github.blog/news-insights/company-news/github-availability-report-march-2026/
10. “GitHub Availability Report: January 2026.” GitHub Blog, February 11, 2026. https://github.blog/news-insights/company-news/github-availability-report-january-2026/
11. “Bringing more transparency to GitHub’s status page.” GitHub Blog, April 17, 2026. https://github.blog/news-insights/company-news/bringing-more-transparency-to-githubs-status-page/
12. “Ask HN: Is GitHub down?” Hacker News, June 29, 2023. https://news.ycombinator.com/item?id=36526267
13. “Summary of the Amazon DynamoDB Service Disruption in the Northern Virginia (US-EAST-1) Region.” AWS, October 2025. https://aws.amazon.com/message/101925/
14. “Google Cloud Incident ow5i3PPK96RduMcb1SsW.” Google Cloud Status, June 12, 2025. https://status.cloud.google.com/incidents/ow5i3PPK96RduMcb1SsW
15. “A Short History of Git.” git-scm.com. https://git-scm.com/book/en/v2/Getting-Started-A-Short-History-of-Git
16. “Git turns 20: A Q&A with Linus Torvalds.” GitHub Blog, June 12, 2025. https://github.blog/open-source/git/git-turns-20-a-qa-with-linus-torvalds/
17. Linus Torvalds, “Re: Kernel SCM saga.” Linux Kernel Mailing List, April 2005. https://lore.kernel.org/lkml/[email protected]/
18. “100,000 Users!” GitHub Blog, July 6, 2009. https://github.blog/news-insights/100-000-users/
19. “Microsoft to acquire GitHub for $7.5 billion.” Microsoft News Center, June 4, 2018. https://news.microsoft.com/2018/06/04/microsoft-to-acquire-github-for-7-5-billion/
20. “Addressing GitHub’s Recent Availability Issues.” GitHub Blog, March 11, 2026. https://github.blog/news-insights/company-news/addressing-githubs-recent-availability-issues-2/
21. “Octoverse 2025: A new developer joins GitHub every second.” GitHub Blog, October 28, 2025. https://github.blog/news-insights/octoverse/octoverse-a-new-developer-joins-github-every-second-as-ai-leads-typescript-to-1/
22. Kyle Daigle, post on X, April 4, 2026, quoted by Simon Willison. https://simonwillison.net/2026/Apr/4/kyle-daigle/
23. Chris Hughes, “The Attack Surface Exponential.” Resilient Cyber, April 14, 2026. https://www.resilientcyber.io/p/the-attack-surface-exponential
24. “AI-Assisted Fake GitHub Repositories.” Trend Micro, March 2025. https://www.trendmicro.com/en_us/research/25/c/ai-assisted-fake-github-repositories.html
25. “Coding on Copilot: Data Shows AI’s Downward Pressure on Code Quality.” GitClear. https://www.gitclear.com/coding_on_copilot_data_shows_ais_downward_pressure_on_code_quality
26. Drew DeVault, “Please stop externalizing your costs directly into my face.” March 17, 2025. https://drewdevault.com/blog/Stop-externalizing-your-costs-on-me/
27. “AI Labyrinth.” Cloudflare Blog. https://blog.cloudflare.com/ai-labyrinth/
28. Daniel Stenberg, “Death by a thousand slops.” July 14, 2025. https://daniel.haxx.se/blog/2025/07/14/death-by-a-thousand-slops/
29. Andrew Kelley, “Migrating from GitHub to Codeberg.” Zig Programming Language, November 26, 2025. https://ziglang.org/news/migrating-from-github-to-codeberg/
30. “GitHub CEO says developers must embrace AI or leave the career behind, citing developer sentiment.” Business Insider, August 2025. https://www.businessinsider.com/github-ceo-developers-embrace-ai-or-get-out-2025-8
31. “Gentoo on Codeberg.” Gentoo Linux, February 16, 2026. https://www.gentoo.org/news/2026/02/16/codeberg.html
32. “GitLab.” https://about.gitlab.com/
33. “SourceHut: The hacker’s forge.” https://sourcehut.org/
34. “Amazon added safeguards including mandatory peer review after AI-tool-caused outages.” The Guardian, February 20, 2026. https://www.theguardian.com/technology/2026/feb/20/amazon-cloud-outages-ai-tools-amazon-web-services-aws
35. “AWS outage tied to Kiro.” Financial Times, 2026. https://www.ft.com/content/00c282de-ed14-4acd-a948-bc8d6bdb339d
36. Dan Whateley, “Replit CEO apologizes after AI coding tool deleted company database.” Business Insider, July 2025. https://www.businessinsider.com/replit-ceo-apologizes-ai-coding-tool-delete-company-database-2025-7
37. Vlad Fedorov, “An update on GitHub availability.” GitHub Blog, April 28, 2026. https://github.blog/news-insights/company-news/an-update-on-github-availability/
38. Mitchell Hashimoto, post on X, April 28, 2026. https://x.com/mitchellh/status/2049213597419774026
39. Mitchell Hashimoto, “Ghostty Is Leaving GitHub.” April 28, 2026. https://mitchellh.com/writing/ghostty-leaving-github