Copilot Lite reviews add 47% more severe fixes

GitHub says Copilot Code Review’s Lite mode now uses an ensemble of agents, increasing the average number of addressed high-severity comments by 47% in experiments while cutting review cost about 8%. The September 11 changelog says medium-severity findings rose 31% and low-severity findings 11%, while shell tools let the reviewer run builds, tests, scripts, and APIs behind its agent firewall. Pilot it on bounded repositories and compare accepted fixes, false positives, and actual spend against GitHub’s earlier HydraFusion workflow-cost analysis before making agent review a default.