Cursor's code review agent can now learn from activity on PRs to self-improve in real time. 78% of issues found are resolved by the time the…

Cursor's code review agent, Bugbot, now supports real-time self-improvement by learning from activity on pull requests. Bugbot reviews hundreds of thousands of PRs per day and uses signals from each review — including developer reactions to its comments, developer replies explaining issues, and comments from human reviewers flagging missed bugs — to generate and refine rules that influence future reviews. Since its beta launch when 52% of identified bugs were resolved before PR merge, the resolution rate has climbed to nearly 80%, which is 15 percentage points higher than the next-closest AI code review product. Bugbot processes these signals into candidate rules that it evaluates against incoming PRs, promoting rules to active status as positive signal accumulates and disabling rules that generate consistent negative signal.

Related

• Learn more: http://cursor.com/blog/bugbot-learning
• Agent responsibly
• New Guide: Incorporating human judgment in the agent improvement loop Building agents is hard. Everyone talks about the code. What gets less…
• Human judgment in the agent improvement loop
• What does it actually take to make agents better over time? A system that starts with a trace. You capture traces of agent behavior, enrich …
• a useful mental model on how teams can think about good data design to improve their models/agents: Evals ~= Training Data ~= Environments -…

Source: agents