VIP 5 min
How to explain AI to people who don’t work in tech
We help people understand and use AI at work. See how we do it

Key takeaways
- Most people don’t need to know how AI works. They need to know where it helps, where it fails and who stays responsible.
- Explain it with a metaphor from the audience’s own world, then say where the metaphor breaks. That is the real lesson.
- Show AI on a real task before any jargon, and write down who checks and who decides.
Very few of us could explain how a car’s gearbox works.
We drive anyway.
What we know is what the car does when we press things, what it can’t do, and what happens if we ignore the warning lights. That is enough, and it’s what we should do when explaining AI at work.
To explain AI to non-technical people, I start from something they already understand. I pick one comparison from their world, say clearly its limitations, and then show it working on their own tasks.
Most people don’t need the engine
Picture the usual “What is AI?” presentation. A diagram of a neural network. Maybe the words “transformer architecture”. Half the room is thinking about lunch, and the other half is thinking: fine, but what does this mean for my job?
This question is the one the presentation should always be answering. Most people need to know where it can help them, the dangers and how to control it.
A finance team does not need a lecture on training data. It needs to know that AI can summarise a long supplier contract in minutes, that it can also miss the clause that matters, and that the person who signs is still the person responsible.
So before I explain any topic, I find out what the audience knows and I draw my presentation from there. My storytelling meets people where they are. People don’t have to fit my story.
A good metaphor meets people where they are
In Learn Like a Pro, Barbara Oakley and Olav Schewe explain why metaphors work. The brain uses something it already knows to understand something new. You are not building a structure from nothing. You are borrowing one.
Their second point matters just as much. No metaphor is perfect. Pick the one that captures the key aspect, then look at where it breaks, because the break shows what the new thing is not.
I can turn this into a formula:
AI is like [metaphor], but [where it breaks].
For example: AI is like a genie in a lamp (the metaphor), but it is only as smart as the request (where it breaks).
With AI, that second step is where most explanations stop too early. “It’s like a very smart assistant” is fine. But leave it there and people will trust it the same way they trust a human assistant. Which in the case of AI is too much.
Three metaphors for AI, and where each one breaks
Here are three examples, and they are just examples. I’m not saying any of these is the right one for your case. The right one is whatever your audience already lives with.
- AI is a fast, well-intentioned intern on their first morning. It can draft, summarise and suggest, and you check the work. Where the metaphor breaks: an intern tells you when they are guessing. AI usually doesn’t. It sounds equally sure when it is right and when it is wrong.
- AI is the predictive text on your phone, but bigger. It feels like magic, it knows exactly the next word. That’s prediction and fluency. But fluent doesn’t mean true. It produces likely next words, it does not look up a fact. Where the metaphor breaks: it can do far more than finish your sentence, so people may underestimate it.
- AI is a search engine that shows you the answer instead of links. People know how to search. Where the metaphor breaks: with a search engine you see the sources and judge them. With AI you often have to ask for them, and then check they exist.
Each “where the metaphor breaks” is the real lesson.
Five steps to explain AI to people who don’t work in tech
- Start from their job. List three tasks your audience does every week. The explanation will live inside those tasks.
- Choose one metaphor from their world. For a maintenance team it might be a new apprentice. For lawyers, a trainee who drafts and never signs. One metaphor well explained is better than five explanations.
- Say where it breaks, out loud. This is the safety briefing. Turn each break into a rule people can remember.
- Show it on a real task before any jargon. Let people watch AI draft something they know well, so they spot what it got wrong. Recognising the mistake teaches more than a warning does.
- Leave them with what is still theirs. Who checks, who decides, who to ask. Write it down where they will see it again.
How we did it: a medical record for a part
We used the same technique on subjects with the same technical complexity, for audiences that thought they didn’t need the training.
A maintenance team was moving from paper records to a digital system. The course did not open with software right away. It explained a log card in one comparison, before a single screen appeared: a log card is like a medical record but for mechanical parts. Everyone in the room knew what a medical record is for, why it must be complete, and why there should only ever be one original. The software then arrived as the sensible way to keep that record.
For export control specialists the problem was the opposite. They already knew a lot, and the subject was a long, precise glossary. We turned it into a TV magic show, where each definition was set up, revealed and explained, with a bake-off in which control language became a pizza menu. The content stayed as precise as the specialists needed. The format gave them a reason to watch to the end.
FAQ
Should I explain how large language models work?
Only as far as it changes what people do. “It predicts likely words, so it can sound right and be wrong” is useful. A diagram of the architecture usually isn’t.
What is the best metaphor for AI?
The one your audience already understands from their own work. Test it with one question: where does it break, and is that break a lesson I want them to learn?
How do I explain AI to people who are afraid of it?
Fear is a separate conversation, and it comes first. I will write about it in a later post on rolling AI out to employees. The short version is to give the reasons before the buttons.
Do people need a course to understand AI?
Not always. A clear explanation, one real task and a short list of rules can go a long way. A course helps when people need practice, or when a rule like the EU AI Act asks you to show their AI literacy.
Back to the gearbox
Most people did not learn to drive by studying the gearbox.
They learned by getting in the car.
More on this: AI at work hub · Seven hundred parts case study
Got a subject people need to understand and act on?
A free first call. You bring the subject, we bring how we'd approach it.
Test us on one subjectTeams we have worked with




