At QCon AI Boston 2026, Roblox's Andrew Swerdlow presented data indicating that autonomous AI agents are outperforming human code reviewers. The AI agents achieved an acceptance rate of 68-70%, compared to approximately 55% for human reviewers. This performance was attributed not to a larger model, but to the AI's training on 700,000 past code reviews, which encoded institutional knowledge into testable rules. AI
IMPACT AI agents trained on institutional knowledge can significantly improve code review acceptance rates, potentially boosting software development efficiency.
RANK_REASON Presentation of new data and findings at an industry conference. [lever_c_demoted from research: ic=1 ai=1.0]
Read on Mastodon — fosstodon.org →
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →