Working with AI takes a teacher’s mindset

I spend a lot of my time teaching people how to use AI at work. Different companies, different teams, very different levels of interest. After enough of those sessions, one thing is hard to miss.

The world's smartest three-year-old

I usually describe a large language model as the world's smartest three-year-old. It knows an absurd amount. It can write, translate, summarise, calculate and code. But it has no idea what you actually want unless you tell it, and it will happily do the wrong thing very well.

So the results depend on the instructions. If you can put into words what you want, who it's for and what a good answer looks like, you get good results. If you type "write me a report on our sales", you get a report on sales. Probably not the one you needed.

Most tools now let you save instructions, background files and preferences, so you don't have to start from zero every time. That helps a lot. But it doesn't change the basic setup: what you ask in the moment still decides how the AI uses everything it's been given. Good notes don't fix a bad question.

What I see in trainings

Most people don't struggle with the tool itself. They struggle with saying what they want. I don't mean that as criticism, because most of us have never had to. At work we lean on shared context: a colleague already knows the client, the history and what the boss likes. The AI knows none of that. You have to hand it over.

In my experience, the people who pick it up fastest are often not the technical ones. They're the people who learn by figuring things out. They push buttons to see what happens, and along the way they learn to explain things in a way the listener understands.

Working with it is teaching

Think about what a good teacher does. They explain the task in a way the student can follow. They give the background the student is missing. They check what came back, point out what's wrong and explain it again a different way. They break a big thing into steps. They're patient, and they're clear about what good looks like.

That is pretty much what working with an LLM is. You're teaching a very fast, very well-read student who knows nothing about your situation.

So I think a teacher's mindset is going to be more and more useful as time goes on. That doesn't take anything away from the engineers. Someone has to run the engine room, and they're very good at it. But the engineer who assembles a Rolls-Royce isn't necessarily the best person to drive one. Building these tools and getting good work out of them are different skills, and the second one is the one most of us will need.

What a good teacher actually has

It isn't only patience and clear explanations. A good teacher has a set of meta-skills: how to figure things out when the answer isn't obvious, how to find another approach when the first one doesn't land, and how to notice when something in the system isn't working, whether that's a lesson plan or a student who has quietly stopped following.

A good manager or salesperson has a lot of this too. What a teacher does on top of it is change how they explain to fit the person in front of them, and check whether that person actually understood. That's the part that maps most closely onto working with an LLM.

Those are the skills that matter with AI. When the answer is off, you need to see it, work out why and try a different way in. Underneath that sits the context of your actual job, because you can only judge the answer if you know what good looks like in your own work. The general skill comes first, and the job knowledge makes it useful.

A side effect I like

There's one effect of all this I didn't expect. To use AI well, you have to communicate clearly. You say what you mean, give the background and be specific about what good looks like. Those are the same things that make us better at understanding each other, and better at saying what we actually want in life. I think that by learning to tell a machine what we want, we might accidentally become better communicators with each other along the way.

Maybe there is some hope after all.