Developer shares regression-check workflow for cleaning up AI-written code

On November 11, 2026, a developer who runs a cleanup service for AI-generated code published a piece on DEV Community describing how he validated the proof that a cleanup did not break anything. The apps he refers to were built with Lovable, Cursor, Bolt or Claude Code. According to the article, such an app ships working and gets used, but afterward every new feature tends to break an old one, and the AI keeps regenerating code it has already produced. The proposed remedy is a cleanup, with the risk that a cleanup can break an app that currently works, and that this may not be discovered until a customer notices. Because his service promises proof that nothing broke, he decided to test whether that proof could hold up. To do so, he ran his own checking workflow against an open source project that ships with 817 automated tests, and he tried to deliberately trick it in order to gauge its reliability. He adds that the scan surfaced the residue typical of a year of AI-written code: the same functionality implemented several times, and stale file copies with no references. Only this one report, dated November 11, 2026, is available, so the exact mechanics of the checking workflow, whether all 817 tests passed, and the name of the open source project used are not stated in the material.