Spotify says verification loops lifted AI agent success rate from 20-30% to 80%

Spotify Chief Architect Niklas Gustavsson said that adding feedback (verification) loops to the company's internal AI coding agent raised the agent's success rate from 20-30% to 80%. The remark, made in a conversation with Boris Cherny, who worked on Claude Code at Anthropic, has drawn attention as a view that the key to running AI agents lies not in the model itself but in the design of the loops.