108 PRs in eight days: Accidentally discovering loop engineering

A developer created an AI agent that completed 108 pull requests in eight days—roughly 10-20 times their previous pace—by automating code writing, testing, and PR management while maintaining human oversight at the testing and deployment stage. The system works effectively on their small, high-trust team that communicates daily, and they discovered this approach aligns with "loop engineering," a methodology for designing systems that guide AI agents rather than relying on manual prompts. They noted this workflow differs from traditional practices by batching releases instead of using continuous deployment, which reduces testing overhead and limits potential impact when issues occur.