After six months without ChatGPT, Copilot Chat, or AI-generated code in production, the author found the experiment changed how they design APIs, debug systems, read unfamiliar code, and write comments. The first week was "embarrassingly slow," with tasks that normally took forty minutes sometimes taking twice as long. The slowdown came from spending more time naming things, checking function signatures, reading package source code, and opening debugger windows. The author felt less capable than remembered. A key example was building a retry mechanism for an HTTP client, which forced questions about whether every error should be retried, whether the server response should be trusted, what happens when the request body cannot be replayed, how cancellation propagates, and whether two retry loops could accidentally multiply the load. The final code was not especially beautiful, but every line had a reason. The experiment began because the author realized that a few days after completing tasks, they often could not explain why a solution was built that way, missing the reasoning in between. The rule allowed documentation, source code, issue trackers, books, debuggers, profilers, and search engines, but not AI assistants.