Spotify more than doubled monthly merged software changes from roughly 8,100 to 17,000 year over year in August as it expanded AI-assisted development. The company is strengthening testing, monitoring and rollback systems because its verification controls did not grow as quickly as change volume.
The scale is substantial: Spotify runs nearly 3,000 production services for 777 million monthly users and handles 11 to 12 million backend requests per second. Its Fleet Management system has merged over 2.5 million automated maintenance pull requests. A newer agent called Honk runs Claude in Kubernetes pods to make more complex changes, while Fleetshift schedules work and tracks progress.
Automation shortened one Java migration across backend services to three days. It also exposed limits in existing checks: an automated dependency update passed validation and then failed in production. Spotify responded by expanding rollback capacity and scheduling automated changes when the responsible teams are working.
The company found no material direct contribution from AI-authored code to its reviewed incidents, and reported no matching increase in rework. However, pull requests and code complexity are growing. The lesson is not that more changes are inherently unsafe, but that deployment safeguards, ownership and observability must scale with the output of coding agents.