Meta Ranking Engineer Agent (REA): Autonomous ML Lifecycle Agent Doubles Model Accuracy Across Six Production Ads Models

Engineering at Meta·high signal

Meta's REA autonomously executes end-to-end ML experimentation for ads ranking — hypothesis generation, training jobs, debugging, and multi-day iteration — via a hibernate-and-wake mechanism with human oversight at strategic checkpoints only. In its first production rollout, REA-driven iterations doubled average model accuracy over baseline across six models. Three engineers using REA delivered launch proposals for eight models, work that previously required two engineers per model.