GitHub created Copilot. Then Copilot broke GitHub.

84.88% uptime over the last 90 days.

Not 99.9. Not 99. Eighty-four point eight eight.

From May 2025 to April 2026: 257 incidents. 48 major outages.

One serious disruption per week. GitHub Actions alone: 57 outages.

Here's why this matters for every engineer shipping in 2026:

1️⃣ GitHub launched Copilot in June 2021

First AI coding tool to hit mainstream adoption. Trained an entire generation of developers to lean on AI for everything.

2️⃣ The market did exactly what GitHub trained it to do

Agentic workflows exploded after December 2025. Repository creation, PRs, API calls, automation all growing at machine speed.

3️⃣ GitHub's own CTO admitted the miscalculation

They planned to scale 10x in October 2025. By February 2026, they realized they needed 30x. They're chasing a curve they helped create.

4️⃣ Mitchell Hashimoto kept a journal

Co-founder of HashiCorp. Marked an X every day a GitHub outage blocked his work. Almost every day had an X.

His 52,000-star Ghostty project is leaving the platform.

5️⃣ Zig left in December 2025

Migrated to Codeberg. Cited a critical Actions bug reported in April that took months to fix.

Freelance dev David Bushell named the pattern:

GitHub is "effectively DDoSing themselves with slop."

That's the framework. The platform isn't being attacked. It's being buried by the automated activity it taught the industry to produce.

Here's why most engineers miss the real story: the failure isn't technical. It's commercial.

Microsoft made Copilot one of its most strategically important products.

When the GitHub CEO stepped down in August 2025, they didn't replace him. They folded GitHub into Microsoft's CoreAI org.

Nobody at the top has GitHub's reliability as their sole mandate anymore.

GitHub's stated priority order now: availability first, capacity second, features third. Right order. Whether they can hold it under Copilot revenue pressure is the question.

The deeper read:

PyPI handles AI scale. npm handles AI scale. GitLab handles AI scale.

They aren't where the agents live.

GitHub is. That's the cost of being the center of the workflow.

Are you still planning your CI/CD around 99.9% uptime that no longer exists? 👇

♻️ Repost for the engineer who lost half a day to an Actions outage this week.