Cursor takes Bugbot out of beta for automated PR reviews

AI can now generate hundreds of lines of code in seconds, but human code review hasn't gotten any faster. If anything, the flood of machine-generated pull requests is drowning senior engineers in review backlogs.

Cursor is tackling that bottleneck directly. The company just officially launched Bugbot out of beta, moving its automated code review agent into general availability.

Originally built to review Cursor's own internal PRs, the tool analyzes code changes to hunt down logic bugs, missed edge cases, and security vulnerabilities before anything touches production.

Catching bugs after humans say yes

Bugbot triggers automatically on pull requests, gathering context across your codebase to parse the real intent behind the changes. Cursor claims it pairs top-tier frontier models with proprietary techniques to minimize annoying false positives.

Teams can also drop a BUGBOT.md file into their repositories to feed Bugbot custom rules and institutional knowledge. When it spots an issue, developers can push it straight into Cursor or launch a background web agent to fix it with one click.

During its beta run, Bugbot reviewed over 1 million PRs and flagged more than 1.5 million issues. Most telling is the follow-through: over 50% of the bugs it caught were resolved before the PR was merged.

Early engineering teams reported reclaiming about 40% of their review time, with several noting that Bugbot frequently caught subtle bugs even after human reviewers had already stamped their approval.

Why it matters

We think Bugbot addresses the biggest emerging problem in modern software development: the review bottleneck created by generative AI.

When AI helps developers write code five times faster, PR volume explodes, and human reviewers naturally start skimming. Having an agent scrutinize AI-written code for logic errors brings the quality control layer up to speed with the generation layer.

For engineering managers and dev teams on Cursor, turning this on in your dashboard is a no-brainer. Catching just one critical production outage before deployment pays for the tooling many times over.

The real risk here is automation complacency. When AI writes the code and AI reviews the diffs, teams can easily slip into rubber-stamping mode. Bugbot seems great at catching nuanced logical slip-ups, but high-level architectural judgment still needs a human in the loop.