Comment by orenp

I'd say the new problem is knowing when understanding is important and where it's okay to delegate.

It's similar to other abstractions in this way, but on a larger scale due to LLM having so many potential applications. And of course, due to the non-determinism.