This is the paradox. It's often harder correcting someone else's work than it is doing it in the first place. So you end up having spent more effort.

It's although a lot less enjoyable in for many. I think the fun stories are all about "look what I have build" and not "look at my amazing code review".

For Boilerplate code we need an AI that is less creative and more secure and predictable. I have fun creative a system design with the right tables and I have implementing logic and interaction design. I don't have the biggest fun writing down dtos and entities.

I would need an AI, that can scan an image and just build me the right lego bricks. We are just getting back to an machine that can do UML from less precise sources.