When I am asked to train students or managers in AI, the temptation is always the same: open a tool and show what it can do. That is exactly what you must not do first. An audience dazzled by a demonstration leaves impressed, and just as dependent. My job is the opposite. To make them able to do without me, and without the tool of the day, because that one will have changed in six months.
I teach in three rather different institutions: the IAE de Corse, where I run a module on AI and data management at Master’s level, the ESM on AI applied to operations at Master’s and Executive level, and the International Business School in Budapest, in English, before an international Master’s and MBA audience. Distant contexts, but the same backbone every time. Understand, use, judge. In that order, and the order is everything.
Understand, first
Before touching a tool, you learn what it really does. Not the mathematical detail, but the principle. A language model predicts the next word from everything it has read. It does not know, it has no store of truth, it produces the plausible. That single idea, well anchored, changes the whole relationship to the machine.
I show it with a real case, because one case is worth a thousand explanations. That New York lawyer who, in 2023, filed six court decisions invented from scratch by ChatGPT, and was sanctioned for it. Once the room has understood why the machine could do that without the slightest hesitation, no one looks at it as an oracle any more. They look at it for what it is: a powerful and fallible tool. That is the start of autonomy.
Understanding is not becoming an engineer. It is ceasing to credit the machine with an intelligence it does not have, and ceasing to deny it the one it does have.
Use, next
Then comes practice, and here I am demanding about the concrete. We do not play with textbook examples. We take a real task from their trade, or their future trade, and we handle it with the tool, from start to finish, with its failures.
In my courses, students actually build. At the IAE, one cohort built its own working assistant in a handful of hours, starting from nothing. What they take from it is not the technical feat, it is what use reveals and no speech can show: the places where the tool saves a huge amount of time, and the ones where it becomes a trap the moment you trust it with your eyes shut. You learn a technology by making it and bumping into its limits, not by listening to it being praised.
This is also when the right reflexes get installed. Check a source rather than believe it. Rephrase a question when the answer falls wide. Recognise a confident but empty answer. These gestures are not taught in theory, they are caught by doing.
Judge, last
This is the step almost every course skips, and it is the most important. Knowing how to use a tool says nothing about knowing when to use it, and when to hold back.
To judge is to decide where the line sits. What I delegate to the machine, what I keep, what I always check. It is also gauging the stakes: on rephrasing an email, a mistake has no consequence, you go ahead. On a decision that commits people or money, you keep your hand on it, whatever it costs in time.
This is where my teaching meets my research on how leaders decide. A manager who knows how to judge does not endure AI, they employ it. They keep responsibility where it cannot be delegated, and they let go where the tool only saves them time. That skill, judgement, is exactly what gains value as AI makes the rest ordinary.
Why I insist on this order
You might think the order is secondary. It is central. Starting with use, as most courses do, produces people who are skilful but without perspective, who use the tool well until the day it deceives them. Starting with understanding lays the foundation that makes use safe and judgement possible.
And then there is a deeper reason. A course is measured by what is left of it a year later. The tools will have changed, the interfaces, the names, the features. What does not expire is having understood what this technology really is, having tested its limits with your own hands, and having learned to decide your place towards it. That is what separates someone who follows the fashion from someone who knows how to use it to decide.
It is also why I never try to turn my audiences into technicians. I try to make them legitimate in front of the tool. Able to ask the right questions, to spot an empty promise, to keep a cool head when everyone gets carried away. The technical side, they will find people for that. The judgement, no one will carry in their place.