I think the risk isn't just that LLMs won't exist, but that they will fail at certain tasks that need to get done. Someone who is highly dependent on prompt engineering and doesn't understand any of the underlying concepts is going to have a bad time with problems they can't prompt their way out of.
This is something I see with other tools. Some people get highly dependent on things like advanced IDE features and don't care to learn how they actually work. That works fine most of the time but if they hit a subtle edge case they are dead in the water until someone else bails them out. In a complicated domain there are always edge cases out there waiting to throw a wrench in things.
This is something I see with other tools. Some people get highly dependent on things like advanced IDE features and don't care to learn how they actually work. That works fine most of the time but if they hit a subtle edge case they are dead in the water until someone else bails them out. In a complicated domain there are always edge cases out there waiting to throw a wrench in things.