Is there really that much skill involved in using LLMs effectively, most of the criticisms I see end up being countered with something along the lines of "you're not using the right model". That implies that much of the skill people talk about is less important than picking the correct model to use (which tend to be the more expensive ones).

And in your LLM future, who will maintain all of the legacy systems that are written in languages the LLMs don't end up assimilating? It's reasonably safe to assume there will be plenty of work left there.