Your team already uses artificial intelligence, and probably uses it well: there are licences, there are results, someone saves hours every week. The uncomfortable question is not whether they use it. It is another one: if tomorrow AI had to take on a serious share of the department's work, who on your team would know how to direct it — ask it for finished work, review what it produces, correct it when it drifts? That capability does not come with the licences. It is trained. And how it is trained is what separates an expense on courses from a capability the company keeps.

A course teaches the tool; the licence is earned by driving

The AI training most commonly offered is the generic course: what a model is, how to write a good instruction, examples from other sectors. The problem is usually not the content but the transfer: the following Monday, the department's real work is still the same as ever, with its data, its deadlines and its exceptions — and the course stays in the certificates folder.

That is why our training follows a different rule: a driving licence, not a course. A licence is not earned by reading the car's manual; it is earned by driving, on real streets, with an instructor alongside. Applied to a department: the training happens on the real work — its documents, its cases, its business rules — and every session ends with something that did not exist before and now works. You do not learn about AI; you learn by directing it at your own desk.

Advanced users, not data scientists

Nor is the goal the one many fear. It is not about turning the team into technicians: it is about training advanced users, not data scientists. An advanced user knows how to do four things an occasional user does not:

  • Ask for finished, verifiable work, not loose answers: a report with its sources, a proposal with its assumptions in plain view.
  • Review before accepting: demand that every figure carries its source, and know how to tell when the machine knows and when it fills in (how that verification circuit is built).
  • Correct where the correction persists: not repeating the same warning every morning, but changing the criterion in the document the system obeys — the difference between using AI and directing it.
  • Delegate with limits: decide what the machine resolves on its own and what passes through a person, and put it in writing.

Behind all four there is a fact worth stating plainly: the system does not improve just by being used. Improvement comes from people who know how to ask, verify and correct. That is why training is not an add-on to the AI project; it is one of its structural parts.

The training that counts is the one that builds something

In our method, the department that operates with AI is built with the client's people, and that construction is the training. There are not two projects —a technical one and a separate course—: the same sessions that leave the system running leave the team knowing how to direct it, because they assembled it with their own hands.

What remains at the end is the difference between the two ways of training. A course leaves a diploma. A construction leaves three things: a working system, a team that knows how to direct it because they watched it come to life, and the documented record of how it was done. The capability stays on the payroll, not on the supplier's invoice — and that changes the nature of the expense: what is rented leaves with the contract; what is trained stays and grows.

The tailwind: the law already asks for it

There is one last argument, and it belongs in the right order: it is not the reason, it is the tailwind. The European AI Act requires whoever provides AI systems and whoever uses them to take measures to ensure, "to their best extent", a sufficient level of AI literacy of their staff — taking into account their technical knowledge, their experience and the context in which the systems are used (Source: Regulation (EU) 2024/1689, art. 4, EUR-Lex, 2024). The obligation has been enforceable since 2 February 2025 (Source: Regulation (EU) 2024/1689, art. 113, EUR-Lex, 2024).

Literacy, in the rule's own definition, does not mean knowing how to program: it is the skills and understanding that allow AI to be used in an informed way, aware of its opportunities and its risks (Source: Regulation (EU) 2024/1689, art. 3.56, EUR-Lex, 2024). The rule does not prescribe a specific course or a certificate: it asks for sufficient capability, in context. Training carried out on the real work and documented session by session produces exactly that evidence — without having worked for the regulator for a single day.

The question to take away

It is not how much AI your company has; it is where the capability to direct it lives. If it lives in your team, it is an asset: it accumulates, it is passed on, and it improves every system it touches. If it lives outside, it is an expense that renews every month and leaves with the contract. The training that counts is the one that moves that capability from the invoice to the payroll — and it is easy to recognise: when it ends, something that did not exist before is up and running, and it was your people who set it in motion.