That would be another LLM or a robot then

I think the material point is that there will be far fewer of those people if everything goes well. You should need fewer people to do any arbitrary task when you are leveraging LLMs.

My own opinion is that people are going to have to become creators. And quickly. You can still create digital products, but you'll need to be a lot more quiet about what you're doing. And you'll need to have a facility for abstract thought to come up with ideas that no one else has yet.

With a few exceptions, using an LLM to perform a useful service is something that almost anyone will be able to do. Therefore these jobs will be not pay well.

There is huge variation in how well people can prompt LLMs, prompt-engineering has many tricks that aren't obvious, that's why there's prompt engineers,

That seems kind of like saying "using Excel to add numbers is something that almost anyone will be able to do" -- true, but the difficult part is (obviously a vast simplification) determining which numbers to add, under what conditions, and to decide what to do based on the result.