Meta's AI Agent Platform Automates Performance Optimization, Recovering Hundreds of Megawatts

Meta's Capacity Efficiency Program has developed a unified AI agent platform that automates both finding and fixing performance issues across its infrastructure. The system encodes domain expertise from senior efficiency engineers into reusable skills, operating through standardized tool interfaces called MCP Tools. This approach addresses two key areas: defense through FBDetect, an in-house regression detection tool that catches thousands of performance regressions weekly, and offense through AI-assisted opportunity resolution that identifies optimization possibilities. The platform has recovered hundreds of megawatts of power, equivalent to powering hundreds of thousands of American homes for a year. Automation compresses approximately 10 hours of manual investigation into 30 minutes, while AI agents can generate ready-to-review pull requests from efficiency opportunities. The system serves over 3 billion users, where even a 0.1% performance regression translates to significant power consumption. New applications built on the same foundation include conversational assistants for efficiency questions, capacity planning agents, and personalized opportunity recommendations. The program operates without proportionally increasing team size while expanding to more product areas every half year.