How Prokeep catches breaking changes and prepares to scale AI code review with CodeRabbit

The team needed a way to scale review without weakening its standards for database migration safety, functional correctness, stability, data integrity, and customer-impacting edge cases.

Solution: Give every GitLab merge request an independent first pass

Prokeep evaluated multiple AI code review tools and selected CodeRabbit largely because it integrates directly with GitLab. The rollout began with a small set of repositories, expanded as the team gained confidence, and is now preparing for broader team-wide adoption.

CodeRabbit reviews changes first. Merge requests still require two human approvals, but those reviewers enter the process after CodeRabbit has already examined the change for bugs, risks, and edge cases.

“We’re already seeing some return on that investment even though we probably aren’t taking advantage of most of the features that are available,” Liz said. “As we are becoming more comfortable with the tool, we’re able to trust it more in its reviews.”

Catching a migration that could have broken production

That trust grew when CodeRabbit raised multiple warnings about a database migration that could break the application. Claude had generated the migration, which made CodeRabbit’s independent review especially valuable.

“I know that Claude is the one that wrote that migration,” Liz said. “Having something come through and say, ‘Hey, don’t do this. You’re going to break production,’ is a massive help.”

The finding gave the team more than a corrected migration. It added confidence that serious errors would receive another independent check before reaching production—even when a human reviewer might overlook them in a busy queue.

Results: High acceptance and real issues caught early

In its first three months, Prokeep acted on 69.8% of CodeRabbit’s review comments. For comments categorized as functional correctness, the acceptance rate reached 79.4%.

CodeRabbit also estimated that it saved 4.29 days of reviewer time during that period. For a team of roughly two dozen engineers, that reclaimed capacity matters as AI tools continue increasing the amount of code entering review.

“It is absolutely catching real issues at review time that humans are not,” Liz said.

The value is not limited to obvious defects. CodeRabbit has also found functional edge cases that could affect customers—exactly the kind of issue that can survive a hurried manual review.

Why Prokeep bought instead of built

Prokeep considered building an internal AI review workflow with GitLab hooks and its own models. The team concluded that the long-term cost would extend far beyond the initial implementation.

“It’s not just building it. It’s maintaining it,” Liz said. “I am more than happy to pay for CodeRabbit to be experts in that space.”

CodeRabbit provides the review experience, dashboard, and configuration in the workflow Prokeep already uses. Developers do not have to operate or monitor a separate internal service.

“Having a tool that just seamlessly integrates is important,” Liz said. “Losing even the little bit that we’re using would absolutely affect our speed to deliver.”

Preparing CodeRabbit for a broader role

Prokeep plans to expand CodeRabbit across more repositories, refine its configuration, and increase developer engagement. The team is also exploring features that can turn findings into commits or new merge requests, reducing the distance between a useful review comment and an implemented fix.

The early results give Prokeep a foundation for scaling AI-assisted development while preserving the independent review layer its engineers need.

“Prokeep is already seeing an improvement in the quality of our code because CodeRabbit is catching things that humans are not,” Liz said.