For two years now I've been hearing the same sentence everywhere: "We've rolled out AI." The licences are handed out, the budget signed off, the announcement made to the executive committee. Then I spend a day with the teams and find something else entirely: a tool almost no one opens, and work that hasn't moved an inch.

It's the most common misunderstanding, and the most expensive: confusing giving access with creating use. Handing out licences changes nothing. Owning a gym membership doesn't make you fit. Between "having an assistant" and "using it to work better" lies a gap that technology, on its own, will never close.

Why does it stay so wide? Two reasons, mainly. First, we tell people to "use AI" without ever telling them what for. Faced with the blank page, they try two or three prompts, come away disappointed, and go back to their old habits : the ones that work. Second, management looks the other way. A manager who never touches these tools can give the finest speech imaginable: no one will follow. Here, example is worth more than any training plan.

So where do you start? With the drudgery, not the demos. The use cases that catch on are never the flashiest, but the most mundane: drafting a meeting summary, roughing out a tender response, getting a first version down. Tasks teams already do, where the payoff shows up in the first week. I'm thinking of a large French group that took the logic all the way: it didn't just open access to an in-house assistant, it turned it into a daily reflex. Tens of thousands of employees genuinely use it, well beyond emails and translations. Nothing magical about it: behind it lies a serious training effort, with more than 50,000 people trained in two years. Use that's useful and measurable sustains itself. Vague use fizzles out within a week.

Management has to step back into the picture. That means managers trained before anyone else, people who practise, who share their own scrappy workarounds, and who own up to getting it wrong in front of their teams. It also means paying the price: you don't "free up" time to explore AI and expect already-packed calendars to absorb it by magic. Learning has a cost. Denying it sabotages the very transformation you claim to want.

Then comes going the distance, and getting specific. General awareness training is a good starting point, but it isn't enough. For work to truly change, training has to reach down into each function – marketing, finance, supply chain, HR – and match people's real day-to-day: you don't equip a buyer the way you equip a product manager. The rest happens through repetition, and peer to peer: team rituals where people swap prompts, a handful of internal champions who spread the good finds. Adoption spreads by contagion, rarely by memo.

In the end, AI at work isn't a technology question. The technology is here: powerful, accessible, and no longer waiting for anyone. The real question is human and managerial: clarity about use cases, leading by example, time, and support. The organisations that succeed won't be the ones that bought the most licences, but the ones that grasped that AI isn't one more tool: it's another way of working together.

And that's the whole difference between a company that talks about AI and one that has actually changed.