CodePal: Snap Built an AI Code Reviewer for the Age of AI-Written Code (eng.snap.com)

Snap has developed an internal AI code review assistant named CodePal to address the challenges arising from the increasing speed of software development driven by AI coding tools. While their pull request (PR) rate surged by 60%, the code review process became a bottleneck, leading to longer merge times and potential technical debt. CodePal aims to alleviate this pressure by providing intelligent, real-time feedback on code, having already reviewed 90% of PRs before they reach human reviewers. Its development in-house stemmed from Snap's unique infrastructure needs and the urgency to rapidly implement a tailored solution. CodePal utilizes a sophisticated two-pass approach, allowing it to deeply understand the repository's symbolic context and cross-repo dependencies. This not only helps it catch complex bugs like logic errors and resource leaks but also enables the tool to provide semantic summaries and auto-generate PR descriptions, which significantly streamline the review process. The iterative Review Loop enhances the thoroughness and consistency of findings, while the tool’s architecture avoids expensive repo clones, making it scalable and efficient. The rapid adoption of CodePal further demonstrates its effectiveness; within a quarter, its usage grew significantly, leading to improved bug detection rates and heightened engineer trust in the tool — a critical development for the AI and software engineering communities.