Every job, at its core, is the same loop. You set an objective, you break it into smaller objectives, and you judge whether the result is good enough. AI is very good at one part of that loop. Give it a clear target and a way to measure progress, and it will optimise for it. What it cannot do is decide what the target should be.
This is easy to miss because we keep saying AI "does the work." In practice it does the middle of the work. The hard parts sit on either side. Defining the problem well enough that a solution actually means something, and deciding what counts as a good answer. Sometimes the objective itself is wrong, and the real work is noticing that before you have spent a month on it.
I see this every day in code. The agents move fast once the goal and the tests are clear. The moment the goal is vague, they produce a lot of confident output that solves the wrong thing. The same holds well outside engineering, in marketing, research, operations. The bottleneck moves from doing to defining.
So the fear that AI replaces us gets the shape slightly wrong. It replaces the execution. The value moves up the chain, towards the people who can frame a problem and recognise a real answer when they see one. That was always the rarer skill. Now it is most of the job.