Why GitHub’s Reliability Issues Are Driving Users Away

GitHub was historically very reliable, but over the past year its normal‑operation time has dropped dramatically. In 2024 the platform recorded 119 service incidents, 26 of them major, with an average resolution time of about 106 minutes. From May 2025 to April 2026 IncidentHub tracked 257 independent incidents, 48 of them classified as major, averaging one major failure per week; February 2025 alone saw 37 incidents, the worst month on record. GitHub Actions was the most affected service, suffering 57 failures in the same 12‑month period, often halting entire engineering workflows.

The official status page reports a 99.79 % uptime, but an unofficial tracker built by Marek Šuppa shows only 84.88 % uptime over the last 90 days, highlighting a discrepancy between reported and actual availability.

According to Šuppa, the unofficial tracker extracts data from GitHub’s own status reports and recomputes service‑level metrics because the composition of GitHub’s average figures is opaque. This transparency reveals the scale of the problem.

Prominent users have publicly expressed frustration. In December 2025, the Zig language project migrated to the nonprofit alternative Codeberg, citing not only security flaws but also a “decaying” engineering culture and a severe GitHub Actions bug that caused build servers to hang indefinitely. Mitchell Hashimoto, co‑founder of HashiCorp, documented daily diary entries of GitHub service interruptions that prevented pull‑request reviews for two hours each day.

GitHub’s leadership has responded quickly, but patience is eroding. The company’s COO replied on X, apologizing and pledging to prove value with evidence rather than rhetoric. CTO Vlad Fedorov later admitted that the platform’s original design cannot handle current AI‑driven workloads, noting that a capacity‑increase plan launched in October 2025 aimed to boost capacity tenfold, yet by February 2026 it became clear that a thirty‑fold increase would be required.

The root cause, as identified by Fedorov, is the explosive growth of code generated by AI agents. Since the rapid acceleration of AI‑assisted development in late 2025, metrics such as repository creation, pull‑request activity, API usage, automation, and large‑code‑base workloads have surged.

Specific incidents illustrate systemic issues. In early 2026 a regression in the merge‑queue feature silently rolled back changes for squash‑merged pull‑request groups, affecting 658 repositories and 2,092 pull requests. The failure stemmed from an unfinished feature flag that allowed new code paths into production without proper monitoring, so the service appeared healthy and no alerts fired.

Four days later, GitHub’s Elasticsearch subsystem collapsed under a botnet‑driven load, rendering most UI content empty and causing widespread disruption.

Commentators such as Hannah Foxwell and Andrew Nesbitt argue that GitHub’s unique role as the hub for AI agents—triggering CI runs, creating pull requests, and interacting with repositories at machine speed—makes it especially vulnerable, unlike other package registries such as PyPI, npm, or GitLab, which have not experienced comparable public outages.

GitHub’s publicly stated priorities are availability first, capacity second, and new features third. While this ordering is sensible, the platform’s ability to maintain it amid relentless AI‑driven traffic and commercial pressure remains uncertain.

In conclusion, GitHub continues to serve over 100 million developers, but the gap between its promised reliability and actual performance is widening, eroding community patience and prompting a search for more resilient alternatives.