article --summary
The summary
This roundup entry describes a test of whether healthier code costs less when AI agents review and fix it. The experiment compared typical unhealthy and healthy pull requests across three language models, 465 pull requests, and 246 open source repositories, scoring code health with CodeScene. The author reports that healthy pull requests came out roughly 22 to 35 percent cheaper to process.
The study also singled out two code smells, Complex Method and Complex Conditional, as drivers of higher review costs. The author takes this as further evidence that code quality still matters in agent-driven engineering.
Editorial note: This is a summary of an article first published by Martin Fowler. The full piece is on their site.