Developers using AI assistants often inherit subtle architectural flaws or dependencies they don't understand. A linter could flag problematic patterns like tightly-coupled modules or fragile prompt-based logic.
The tool would analyze commit histories to distinguish human-written from AI-generated code, then assess maintainability risks. Engineering managers would pay to prevent future rewrite costs.
Start with a GitHub Action that scores pull requests. The challenge is balancing false positives with actionable insights—overly strict rules will be ignored.