You still have to know things.

Without that knowledge, I might have trusted it. The answer alone would not have told me what was wrong.

That is a small example of a much bigger issue. AI gives us access to answers in fields we know very little about. It does not automatically give us the expertise to judge them.

Access is not expertise.

A manager without a commercial background told me he could now make all the commercial decisions himself. With AI, he no longer needed the commercial or finance teams.

I understand the appeal. You can ask a question and get an analysis in seconds. Ask another and get a recommendation. Work that used to require several people suddenly looks like something you can do alone.

But could he tell whether the commercial logic made sense? Whether the financial assumptions were realistic? Whether a crucial cost was missing? Or whether AI had simply made up the numbers?

I am not saying that happened in his case. I am asking how he would know.

Alex M H Smith puts it in one question: do you know what good looks like? That is what expertise gives you. Not just the ability to produce an answer, but the ability to question one. To spot the missing assumption. To know when an apparently sensible recommendation does not fit the situation.

Know what sits behind the answer.

That judgement also applies to the system itself. How do you store your data? What can each agent or chat see? What do you ask it to do, and which decisions do you keep with people?

An answer can accurately summarise the information available and still miss the point because the important information was never there. An error in one step can become the input for the next. A bias in your sources can survive all the way to a polished recommendation.

So you need both kinds of understanding. Enough knowledge of the subject to judge the answer. Enough knowledge of the setup to see where that answer came from and what it might be missing.

Through experts, not around them.

This is a plea for expertise. Not for preserving every task or defending every job title. AI can take over parts of the work. It can also help us learn things we did not know before.

But getting an answer is not the same as understanding it. The commercial and finance teams are not valuable only because they can prepare an analysis. They are valuable because they know which assumptions to challenge and what the consequences might be.

Use AI to extend that expertise, not to bypass it. Let it help you explore, draft and question. Match the level of checking to what is at stake. And when you cannot judge the output yourself, involve someone who can.

You do not have to know everything. You still have to know enough to recognise when you do not.

Source and inspiration: Alex M H Smith on knowing what good looks like, and Dan Hockenmaier on using AI as an input rather than an output.