🔗 Test coverage won't save you

What happens when coding agents take over? In her insightful article, Jenn Cooper highlights a major flaw in relying solely on testing in AI-generated codebases. While high test coverage may seem like a safety net, it primarily locks in existing behaviors, good or bad. This can lead to a creeping architectural drift, where problematic patterns become entrenched without anyone noticing.

To counter this, she advocates for fitness functions that evaluate the overall shape and coherence of the system, not just individual components. By implementing macro-level checks and creating friction during coding, teams can guide agents towards healthier coding practices and ensure a more coherent codebase. The discussion resonates with anyone tasked with maintaining code quality in an era increasingly influenced by AI.