What does expertise become when AI can increasingly perform the execution?

AI has made it possible to do things I could not previously have done unaided.

Coding makes that particularly visible. I can now build software whose implementation is beyond what I could independently produce line by line.

That creates an interesting question. If the AI can do more of the work than I can, where does my capability sit?

One answer would be that expertise matters less. I think something more complicated is happening.

The work is separating. There is the ability to perform an individual task, and there is the ability to understand what task should be performed, how it relates to the wider problem, whether the result is plausible, what should be challenged and what should happen next.

AI is rapidly changing the first. It may make the second more important.

Expertise isn’t disappearing. It’s moving up a level.

That does not mean underlying knowledge becomes irrelevant. Quite the opposite.

The less directly involved someone is in producing every part of an artefact, the more important it can become to recognise when the artefact is wrong.

A model can generate code. Someone still needs to know whether the architecture makes sense, whether the security assumptions are acceptable, whether an edge case matters, whether the tests are meaningful and whether the thing being built actually solves the problem.

The same pattern extends beyond software.

AI can draft an argument, analyse information, create a plan, summarise evidence or propose a decision. The human contribution increasingly includes deciding what to ask for, supplying context, decomposing the problem, challenging the answer, combining competing considerations and taking responsibility for the result.

That is not merely prompt engineering. Prompting is one interface through which direction happens. The more durable capability is directing work.

There is a connection here to clarity.

When execution is expensive, some ambiguity can be resolved during the doing. People encounter the problem, discover what was meant and adjust.

If execution becomes almost immediate, unclear intent can produce the wrong thing at extraordinary speed. The quality of direction therefore matters more, not less.

There is also a tension with apprenticeship.

If expertise moves towards judgement, people still need a way to acquire that judgement. Historically, much of it came from performing the lower-level work repeatedly: making mistakes, seeing edge cases and learning why apparently sensible approaches fail.

AI can remove some of that work at exactly the moment when the judgement developed through it becomes more valuable.

So there are really two questions.

What does expertise look like when AI does more of the execution?

And how do people develop that expertise if they no longer travel through the same path to acquire it?

The first question is optimistic. AI can expand what capable people are able to do.

The second is an organisational design problem we still need to solve.