I think it's a bit like `rails generate`, where it massively speeds up getting a CRUD webapp 0 to 1, but once you get to GitHub or Shopify size, you need a lot more than that to add a new data model.

AIs are pushing many things forward, but due to training sets and context windows, I think meaningfully adding to actually valuable apps, at least as we currently write them (the kind with many DBs/caches/message queues, services) will take a fair bit longer.