▶ YouTube Summaries

• https://www.youtube.com/watch?v=t-DM44d-qhU
• Original title: GitHub is Falling Apart

A heavy GitHub user's critique of GitHub's deteriorating reliability under Microsoft: real uptime has slid to roughly 87% versus the 99.9% SLA, capped by the April 27 outage (pull requests and issues vanishing due to an overloaded Elasticsearch cluster, likely a botnet attack) and an April 23 merge-queue bug that silently reverted code on customers' main branches. He pins the root cause less on AI-driven load growth and more on a leadership vacuum — no CEO since the last one stepped down, with GitHub run as a Microsoft feature rather than a standalone platform — plus misallocated focus on Copilot-branded AI features, a fake-star economy, and a dishonest status page. High-profile users like Ghostty's Mitchell Hashimoto are already leaving.

The April 27 outage and the April 23 merge-queue regression

Two incidents anchor the video. On April 27 users across X reported pull requests and issues failing to load ("failed to load issues", PR lists empty despite counters showing them). The postmortem attributed it to the Elasticsearch subsystem powering search-backed experiences (PRs, issues, projects) becoming overloaded — likely a botnet attack — and ceasing to return results; no data loss occurred.

Four days earlier, on April 23, merge queue produced incorrect merge commits when a merge group contained more than one PR merged via squash: instead of applying only the PR's changes onto the tip of main, it effectively merged back to the branch's base point, silently deleting intermediate commits on main. GitHub's COO Kyle downplayed it as 2,804 affected PRs out of 4M+ merged that day (~0.07%), which the author finds tone-deaf — and notes the official status page shows no incident for Git operations on April 23 at all, despite the postmortem. He also points out the official status page has no aggregate uptime number (individual components all conveniently average above 99%), forcing users to rely on an unofficial GitHub status page that shows ~87.25%.

The leadership vacuum

The core thesis: after the CEO stepped down, Microsoft never backfilled the position. GitHub now has a chief of staff, COO, CTO, two CPOs, CRO, CFO — but no CEO; the COO reports into Microsoft. Microsoft treats GitHub as a division/feature rather than a standalone platform. His analogy: the best crew in every role still runs aground without a captain. Meanwhile people are moving to GitLab or Codeberg, or hoping for something new — Depot (the sponsor) hints publicly it may be building a GitHub alternative, and its Depot CI already replaces GitHub Actions (example given: a 6-minute Actions job dropping to 35 seconds).

Fake stars and misplaced priorities

Citing the "Inside GitHub's fake star economy" article (6 million fake stars, ~6 cents per click), he explains stars are now bought as fake VC traction signals. The article's methodology: compare stargazer profiles of mature repos (Flask, LangChain, AutoGPT — 6-12% of stargazers with zero followers, 2-6% with zero public repos) against manipulated blockchain repos, where 28-38% of stargazers have zero public repos and 50-81% have zero followers — ghost accounts. He ties this to a broader pattern: GitHub shipped Copilot Workspace, Copilot Agents, Spark, Models, Copilot Code Review, Copilot in the CLI in 18 months while the core product experience degraded.

Is AI-driven load a valid excuse?

GitHub cites record activity: 1 billion commits in all of 2025 versus 275 million per week now; Actions minutes growing from 500M to 2.1B per week; 90M PR merges, 1.4B commits, 20M new repos per month (though he mocks the axis-less charts). He grants some leeway for unprecedented load — much of it from agentic coding (e.g. the OpenClaw author committing thousands of times daily) — but notes uptime has been declining since the Microsoft acquisition, before the LLM boom, so AI can't be the whole story. Quoting a tweet: it's hard to sympathize with the 900-lb gorilla — as market leader, either handle the load or lose customers to someone who can.

Ghostty leaves GitHub

Mitchell Hashimoto (GitHub user #1299, joined February 2008) announced Ghostty is moving off GitHub. He kept a journal marking every day a GitHub outage impacted his work — nearly every day had an X — and wrote the post while unable to review PRs for two hours due to an Actions outage (a different outage than April 27). His summary resonates: "I want to get work done and it doesn't want me to get work done. I want to ship software and it doesn't want me to ship software." His personal projects stay, but Ghostty and its team are leaving — and the author suspects Ghostty won't be the last.

Takeaways

• Real-world GitHub uptime (~87-90%) is far below the 99.9% SLA; the official status page obscures this by omitting aggregate uptime.
• The merge-queue incident is the scariest kind of failure: silent data corruption on main branches, not mere downtime.
• The author's stance is constructive, not a dunk: he wants a CEO hired, transparency restored, and the core product prioritized over AI feature sprawl.
• Alternatives (GitLab, Codeberg, whatever Depot may build) are now serious options; watch for more high-profile departures.