I did this for a while… and until Opus 4.5, I couldn't fully trust the model. But at this point, while it does make the occasional mistake, I don't need to scrutinize every line. Unit and integration tests catch the bugs we can imagine, and the bugs we can't imagine take us by surprise, which is how it has always been.