Half your team is quietly terrified of AI. The other half won't shut up about it. Your job is to get them all in the same room — without a fight.

That's the real challenge. The conversation about AI in most businesses is about tools: which model, which subscription, which platform the sales rep demoed last Tuesday. But the tools were never the hard bit. Getting a team of actual human beings (with mortgages, pride, habits, and a perfectly reasonable fear of being made redundant) to change how they work: that's the hard bit. And you don't fix a people problem by buying more software.

Training your team on AI isn't really a training problem. It's a change-management problem wearing a training problem's clothes. Once you see it that way, most of what usually goes wrong starts to make sense, and most of the fixes get a good deal simpler. It's the same point I've made before about why AI maturity is a people story, not a technology story, though here I want to talk about what you actually do on the Monday morning.

The cast you already have

Walk into any team and you'll meet the same three people.

There's the enthusiast, who found ChatGPT eighteen months ago and now has opinions about five tools you've never heard of. There's the cautious middle, the bulk of your team, who aren't against it, exactly, but haven't seen a good enough reason to change a workflow that already works. And there's the veteran, the one who's really good at the job, who looks at all this and sees either a threat to their craft or a fad that'll pass if they keep their head down.

None of them is wrong. The enthusiast is right that this matters. The cautious middle are right to want proof. The veteran is right that a lot of it's hype, and right to be protective of skills that took years to build. Treat any of them as a problem to be managed and you've lost them. (For a fuller map of who's on this journey and what each group needs, I wrote a practical guide to the whole spectrum last year.) The goal was never to convert the sceptics, but to build something that lets each of them move at their own speed without anyone feeling stupid.

Why the training usually misses

This is where most efforts come apart. A company decides to "do AI training", books a session, and someone stands at the front explaining how ChatGPT works — prompts, tokens, maybe a slide about hallucinations. Everyone nods, goes back to their desk, and nothing changes.

The session failed because it taught the tool instead of the job. Nobody in your team wants to know how a large language model works, any more than they wanted to know how the printer works. They want to know how this thing fits into their actual Tuesday morning — the report they dread writing, the inbox that's out of control, the quote that takes an hour it shouldn't. Teach the tool and you get polite nods. Show someone the specific horrible task on their own desk getting visibly shorter, and you get a convert.

Start with something boring

Which is exactly why your first AI project should be dull.

There's a temptation to launch with something ambitious and visible — reinvent the customer journey, automate the whole sales pipeline. Resist it. Pick instead the low-stakes, high-annoyance task that everybody in the room secretly resents. Writing up meeting notes. Formatting the monthly figures. Drafting the same six emails you send every week. Something where the cost of getting it wrong is roughly nothing and the relief of getting it right is immediate. Pick something visible, too — a task enough of the team shares that the win doesn't stay hidden on one person's screen. And you don't need a mandate or a budget line to start: one boring task and two willing people is enough, whoever you happen to be in the org chart.

Do that, and something useful happens. People see a result before you've asked them to change anything important. The enthusiast gets to shine, the cautious middle get their proof, and the veteran gets to watch from a safe distance without having to stake their reputation on it. You're just letting them see the thing work on a job nobody cares about losing. Belief comes after, on its own, once they've seen it for themselves.

Make it safe to be rubbish at first

None of this works if people are frightened, and a lot of them are — of two things in particular.

The first is looking foolish. AI tools are genuinely strange to use well. The first prompts are clumsy, the results are mixed, and for anyone who's spent a career being the competent one, that early incompetence is deeply uncomfortable. So they avoid it, in private, and tell you later that they "didn't have time". You've got to make it explicit, out loud and more than once, that fumbling around is the normal state, that everyone's results are bad at first, and that asking a colleague for help is exactly what you want to see rather than a sign of weakness.

The second fear is the big one, and pretending it isn't there fools no one. People are worried the point of all this is to need fewer of them. If that really is the plan, no amount of training will paper over it, and you shouldn't try. But if it isn't — if you're doing this to get more out of the people you have, not to have fewer of them — then say so plainly, and then behave as though it's true. The job is being augmented, not replaced. That sentence is just a poster on the wall until your actions back it up.

Let it spread through people, not memos

Adoption arrives through the people who already trust each other. You can rarely mandate it into being.

The single most effective thing I've seen is almost embarrassingly simple: pair your enthusiasts with your sceptics. Not in a training session — just as a standing arrangement, an AI buddy the sceptic can lean on. When the veteran hits a wall, they don't have to raise their hand in a workshop and admit they're lost. They turn to someone they already trust and say, "how would you do this?" The enthusiast, meanwhile, finally has somewhere to put all that energy that isn't annoying everyone. I'm not sure why the pairing works as well as it does — my best guess is that people will admit ignorance to one friend at the next desk long before they'll admit it to a room.

That's how change actually moves through a team — sideways, between people who like each other, one small win at a time. A memo from the top telling everyone to use the new tool by Friday does the opposite. It turns a genuine opportunity into a compliance exercise, and people comply with exactly as much enthusiasm as compliance deserves.

Measure what you actually wanted

Be careful what you count, because you'll get more of it.

Most people reach for usage — how many people logged in, how many prompts were fired off. It's an easy number to pull, and it goes up and to the right, which looks like progress on a slide. It tells you almost nothing. A team can be using AI constantly and producing exactly the same work, just with extra steps, the way so many are right now — asking the tool, copying the answer out, pasting it back into the same old process by hand.

Measure the outcome instead, and there are really only three things worth counting: time saved, quality, and problems solved. Did the work get faster — real hours back, not just a nicer-looking interface? Is the quality holding up, or better? Has a problem that used to eat half of someone's week simply stopped happening? If the report, the inbox and the quote aren't better than they were, the adoption isn't working, however busy the dashboard looks. And if they are, you often don't need a dashboard at all. You can hear it in how people talk about their week.

If the boss won't use it, no one will

Which brings it back to you.

If you're asking your team to feel foolish in public, to change habits that work, to be beginners again — you've got to be visibly doing it too. Not a polished demo where you've rehearsed the prompt. The real thing: show them the prompt that didn't work, the draft that came back wrong, the afternoon you lost down a blind alley — and then the version that finally saved you two hours on something you'd have grumbled about for years. Share the fumbles as readily as the wins. A team will forgive a leader who's clearly figuring it out alongside them. What they won't do is take seriously a change the person at the top can't be bothered to make themselves.

None of this is quick, and I won't pretend it's a formula. But the businesses that get this right won't be the ones who bought the cleverest tools. They'll be the ones who remembered that behind every login is a person wondering what it means for them, and who took the time to answer, kindly, before asking anyone to change a thing.