Meta AI Presents CORAL Harness for Recommenders

Meta AI researchers introduced CORAL, an LLM-native harness that deploys an agent in a continual closed loop over a live production recommender system. The approach aims to handle ongoing optimization as content, user behavior, and models change. Posts from AI researchers highlight the paper's demonstration of agent use in a production setting with reported A/B test results. The work focuses on recommender systems that influence what billions of people see daily.