My experience matches yours. Lots of apps are quite repetitive: for building APIs for example you generate one controller and the ask the app to generate more

My experience matches yours. Lots of apps are quite repetitive: for building APIs for example you generate one controller and the ask the app to generate more using the first ones as a pattern. For frontend you do the same for forms or lists. Tests are often quite good, but I think they were already great even back in the first ChatGPT release. With this strategy and the fact that some patterns are quite verbose (albeit understandable for an AI or a reader), it is quite easy to get to a big LoC while still maintaining consistency.