AI code looks clean and is architecturally rotten: the data proves it
AI-generated code arrives syntactically perfect and structurally incoherent: models can't see your codebase's conventions, so they re-implement what utils already do and ignore bounded contexts you spent weeks establishing. The numbers make it concrete — GitClear shows 4× more duplication, refactoring collapsing from 25% to under 10%, and 7.9% churn within two weeks. Passing tests and clean lint are the trap; the debt surfaces in review, months later.