you mean AI research jobs? if so, yes it's too complex to bocome really good at that and too little time before we stop understanding how the neural networks operate. Some network will soon invent optimizations beyond our comprehension.

then they will start design HW and automation to build it. And they will move to supply chains.

I think there's a good opportunity now (may be next couple of years) as consultant for agentic AIs implementation in companies.

then this job will get replaced by agents too.