Does code quality still matter when AI writes the code?

Why it matters

A developer at DocSpring, a document-generation service, ran a retrospective case study to test a popular claim: that AI coding agents make technical debt irrelevant because they can work through messy code. The author screened recent coding-agent transcripts and pull-request discussions, inspected changes, and ran small experiments. The findings push back on the hype. The clearest failures came from behavior duplicated across multiple places and from checks that failed to express the developer's actual intent, not from long files. In one reproduced example, a fix to JSON Pointer path handling for field names containing a slash missed a separate branch governing required-field rules. Another case showed that strict typing in Sorbet only caught a misspelled method override when the signature explicitly declared it as an override. The author reports that Claude Code and Codex help investigate the long-lived codebase, but concludes that consolidation of logic and intent-revealing checks still matter, and every suggested improvement still requires human review and testing.