The Autonomous Performance Agent: A Netflix Production Story

At Netflix, with large codebases and distributed systems, software performance degradation is often silent and compounds without active engineering involvement. The manual cost of closing the loop (profile, analyze, trace, fix, validate) means many inefficiencies quietly burn compute for months before anyone acts. By the time a human gets there, it's often already cost real money. We wanted to see if the loop could close on its own. I'll talk about how to build an autonomous agent that hunts for performance inefficiencies across live production services, traces them to source code, proposes fixes, and validates results — grounding every decision in measured production outcomes, not model confidence. In this talk, we'll share how engineering leaders can scale their performance engineering with newer approaches that give agents enough autonomy while keeping humans focused on the key details that matter. We'll walk through four findings the agent surfaced — caught, fixed, and canary-confirmed — with limited hand-holding from an engineer.