Expertise Isn't Disappearing. It's Moving Up a Level.

There is something slightly strange about the way I code now.

AI can produce software that I could not independently write at the same level or speed. Yet I can still build increasingly sophisticated things with it.

So what, exactly, am I getting better at?

Some of it is technical knowledge. The more I build, the more I understand the systems I’m working with.

But a lot of the improvement feels like something else.

I am getting better at describing what I actually want, breaking a problem into useful pieces, giving the model the right context, spotting when it has gone in the wrong direction, asking it to test assumptions and deciding when an answer that looks convincing isn’t good enough.

In other words, less of the value is in typing the implementation and more is in directing the work.

I don’t think this means expertise is disappearing.

It may be moving up a level.

That distinction matters beyond coding.

AI can already draft, analyse, summarise, research, plan and create. As those capabilities improve, knowing how to perform every step personally may become less important in some contexts.

But knowing what good looks like does not.

Someone still needs to understand the problem.

Someone needs to recognise the missing assumption.

Someone needs to notice that the apparently excellent answer solves the wrong thing.

Someone needs to know when to challenge the model, when to change direction and when AI should not be doing the work at all.

That is more than prompt engineering.

Prompts are just one way of expressing direction. The deeper capability is being able to direct work well.

There is an important catch.

A lot of judgement was historically acquired by doing the work that AI can now help us skip.

You learn what good code looks like partly by writing bad code. You learn where projects fail by experiencing projects. You develop judgement by seeing enough apparently sensible decisions produce unexpected consequences.

So if expertise is moving towards judgement, we also have to think seriously about how people acquire that judgement.

That connects directly to the apprenticeship problem.

AI can make experienced people dramatically more capable. But if it removes too much of the path through which inexperienced people become experienced, organisations may eventually discover that the scarce thing isn’t execution at all.

It is people who know what to ask for, what to believe and what to do next.

Expertise isn’t disappearing.

But we may need to become much more deliberate about where it now lives and how people learn it.