Skip to content
Marc Allington
Marc Allington
← All signals
Agentic engineeringgood news for the people this stream is for

Marcus Borg of CodeScene, quoted on Martin Fowler's site, ran a review-and-fix pipeline of five agents and three models over 465 pull requests from 246 open source repositories. Pull requests in code scoring 2.8 on CodeHealth cost roughly 22 to 35 percent more to process than those scoring 9.7. Complex methods and complex conditionals drove most of the extra cost. For leads, it gives a cost argument for paying down code health before pointing agents at a codebase.