Automating Low-Risk Code Review at Meta: RADAR, Risk Calibration, and Review Efficiency

Meta has introduced RADAR (Risk Aware Diff Auto Review), an AI-assisted code review system that utilizes a multi-stage funnel to automate the review process for software changes. RADAR employs a machine-learned Diff Risk Score and LLM-based automated reviews, having successfully reviewed over 535,000 diffs with a significantly lower revert rate and production incident rate compared to traditional reviews. This system enhances review efficiency, reducing median time to close by over 330% and median review wall time by 35%, addressing the growing disparity between code production and reviewer capacity while maintaining safety standards.