I Built an AI PR Reviewer That Catches Bugs by Not Looking for Bugs (tessl.io)

A new AI-powered plugin, developed as a Tessl module, aims to enhance the pull request (PR) review process by sidestepping traditional bug-finding methods. Instead of searching for bugs, the plugin constructs a dossier of evidence regarding code changes, classifies risks into categories (green, yellow, red), and provides a structured brief to human reviewers. With an impressive accuracy of 97.7%, it emphasizes human judgment by directing attention to significant areas and risks, rather than overwhelming developers with irrelevant or erroneous AI findings. This approach addresses the current dissatisfaction in PR reviews, where human reviewers often feel discouraged due to the imbalance between AI-generated code speed and the extensive time required for thorough inspection. This plugin not only streamlines the review process, allowing developers to focus on intent and architectural decisions that AI cannot assess, but it also sets out to create a feedback loop for continuous improvement through retrospective evaluations. By integrating a second model for cross-verification and synthesizing findings into clear actionable items, the tool represents a significant step towards a more effective collaboration between human reviewers and AI systems. Ultimately, it aims to restore developer trust in AI-assisted code reviews by ensuring that the human in the loop has the most relevant information at their fingertips.