GitHub's Decay: AI Push, Broken SLAs, and a Frontend Falling Apart

Efron Licht, a distributed systems engineer, argues that GitHub’s recent reliability collapse is symptomatic of deeper rot rather than a passing capacity crunch. Despite public claims of 99.8% uptime, the service suffers dozens of self-reported incidents per month, hides issues in email chains rather than a public bug tracker, and has no ‘performance’ or ‘reliability’ category in its changelog at all. Over the past 30 days, GitHub’s patch notes mention ‘copilot’ 59 times and ‘agent’ 8 times while mentioning performance and reliability zero times each — a clear inversion of Microsoft’s stated ‘availability first’ priority.

Licht’s sharpest point is that the ‘agentic workload’ Microsoft blames for the strain is one Microsoft itself manufactured. GitHub subsidized Copilot adoption for years and now stuffs four separate agent-launch buttons into the top-right corner of a typical repo page, effectively DDoSing itself with traffic it created. Meanwhile the frontend is bloated enough to spike browser heaps past 512 MiB on PR review pages, regularly breaks on Firefox and Safari, resets user settings to dangerous defaults, and tolerates open star-buying marketplaces that Carnegie Mellon research has tied to malware distribution.

The broader claim is that GitHub has become a case study in big-tech infrastructural decay: a service so synonymous with software development that recruiters treat its absence as disqualifying, yet one whose owners have chosen flashy AI funnels over the boring work of keeping pull requests, Actions, and the status page honest.