When agents write the code, what must developers still know?
A developer describes Codex writing complex components on single lines, tripping SonarQube's complexity check, then fixing it by refactoring into ten files. The incident frames a broader question: as agents produce more code, how should developers' understanding and ownership of systems change.