The "a ton to learn about how to control an LLM" is remarkably similar to learning how to work with a new team of humans, perhaps easier. LLMs though eccentric are far more predictable than most humans.

What I love most about AI development is the lack of pushback when you see a completed new feature, realise it was a dumb idea and just pivot hard to something different.

A human would (rightfully) call you out for wasting time and resources. AI slurps up your feedback and dives in again with the same vigour.