One GitHub failure halted Actions and two coding-agent tools
On July 9, could not obtain runners for some workloads, and the disruption lasted 10 hours and 10 minutes. About 8% of workflow runs were delayed by more than five minutes, while roughly 2% never started. An internal data service that prepares runners became unhealthy and could not recover on its own, so people had to intervene.
Copilot Cloud Agent and Copilot Code Review also stopped because they depended on the same service. recorded 57 service incidents during the 12 months ending in April 2026, more than any other GitHub service. By early 2026, GitHub was processing 275 million commits each week, compared with about one billion a year previously, and most of the increase was attributed to agents.
Claude Code alone was producing 2.6 million commits per week. Because agents can submit code, open review requests, and launch automated checks around the clock, pressure on this shared may persist.
Key points
- The disruption lasted 10 hours and 10 minutes.
- About 8% of runs were delayed over five minutes, and about 2% did not start.
- One internal service failure affected and two Copilot tools.
- The service could not recover automatically and required manual intervention.
- Agent systems need retry limits and duplicate-job protection to avoid extra costs during outages.