Agreed. But I think we may reach a point where it is difficult to prompt the most advance AI (which may are may note be an LLM, though maybe contain an LLM mode

Agreed. But I think we may reach a point where it is difficult to prompt the most advance AI (which may are may note be an LLM, though maybe contain an LLM mode) in a productive way, especially due to computational demands on resources. And so AI's with lesser, cheaper capabilities may be reasonably competent in "understanding"-- a term I user very loosely-- the problem of dealing with more advanced but resource constrained systems and assist in the best practices in prompting them.