> understanding it from writing it

On the surface it sounds weird - why would this be?

Possibly because building a system is not a one-shot step, but a process of many iterations, each of which involves experiments in production, and gaining more learnings. So at the end of the process, you don't just have N lines of working code, but also N lessons learned along the way. So presumably with the AI process we miss out on half the value.

Now the going thesis is that this extra value is unnecessary if we take the plunge and don't look back. My gut says the answer is somewhere halfway, I guess we'll see.