I'm starting to believe that LLMs have great potential to keep us dumb and unproductive. It's great fun to throw Sonnet at some big code base and get the resemblance of new features in seconds but the more involved it gets the more it tends to introduce errors and then you are just debugging which is a great opportunity to add more errors and then the task can only be abandoned after spending time and money on it.

So if you have no idea about coding, it gives you a magic wand that can summon stuff you couldn't imagine before but for somebody who has summoned stuff for many years this magic wand is kind of under-powered still.