The article discusses how AI changes the landscape of code reviews, making the reviewer's job more complex. It outlines specific heuristics for assessing pull requests (PRs), focusing on aspects like design, testing, error handling, and the effort put in by the author. The author emphasizes the need for human oversight despite advances in AI review tools.

The author argues against traditional line-by-line code review, advocating for a harness-first approach where specifications and testing take priority. They draw on examples from AI-assisted coding and highlight the importance of architecture and feedback loops over direct code inspection. Caveats are noted for critical systems where code review remains essential.

This article outlines how Qodo developed a benchmark to evaluate AI code review systems. It highlights a new methodology that injects defects into real pull requests to assess both bug detection and code quality, demonstrating superior results compared to other platforms.

The article stresses the importance of software engineers providing code that they have manually and automatically tested before submission. It emphasizes accountability in code reviews and the use of coding agents to assist in proving code functionality. Developers should include evidence of their tests to respect their colleagues' time and efforts.