Stop Sending IDE-Catchable AI Code Errors to Review | The JetBrains AI Blog

The rise of AI coding tools has significantly increased the volume of code arriving for review, presenting new challenges for engineering leaders as traditional code review processes struggle to keep up. With developers using AI tools in an ad hoc manner, these tools have introduced unique error patterns and a distinct error profile, including unused constructs and higher-risk security vulnerabilities, which are not common in human-written code. This influx of AI-generated code has led to a 7.2% reduction in delivery stability as it places a greater burden on human reviewers who must now contend with larger changesets and more complex error profiles. Automated structural and static analysis before code reaches human reviewers can alleviate some of this burden, as evidenced by companies like Google and Uber, which have implemented automated verification systems to reduce the need for human intervention and maintain stability. While many development environments rely on language-by-language approximations, comprehensive structural analysis at an organizational level is crucial for maintaining code quality, with tools like JetBrains IDEs and Qodana offering solutions for pre- and post-pipeline checks to protect reviewers' capacity and judgment.