While I do agree, if we switch the role from getting results to training the model, the job of engineering the right prompt to get the right result is simply called an annotator (which is how AIs are currently refined).

So I'd say it would not be a short lived job, just a multi purpose one at first, slowly switching to model training over time. Just like how a lot of data scientists have slowly switched from model research to data cleaning. (the above is indeed a personal opinion, not a prediction)

More concisely, I think any job that relates to training/refine AIs, may it be by deeply annonating or extensively using it, is a job of the future.

example : a picture generation AI service, hiring an expert in Art History from the romanticism era to create high quality skills in that domain, with the right prompts and corrections in the shortest time. An AI teacher, to put it more simply.

example 2 : a very highly ranked competitive FPS gamer to play extensively with an AI, which would be included as a coop bot in an upcoming FPS.

The choice of the teacher could even be part of the company branding : Company A capitalizing on offering bots or NPCs "inspired by the famous ProGamer69 style", Company B "inspired by the even more famous Crusher69". etc

Right now we are at the infancy of AI creation, but I think we'll soon enter into the identity crafting era, with several products declining the same AI model source, upon different pre-trained sub-expertises. This will be very interesting.