In brief. Companies name skills as the number one brake on AI, ahead of cost and technology (Istat 2025). But a one-off course for the whole team has the same flaw as customer onboarding without tracking: nobody really knows who learned something they use every day and who just attended. What works better is an internal reference point per department, the AI Champion, who practises on real processes and spreads what they learn. The skills that count are not technical but practical: delegating the right tasks, writing clear instructions, recognising when not to trust the output.
"We did the AI course, now we know how to use it"
It is the sentence I hear most often when I walk into a company that has already tried to move on AI. The team did two or three days of training, everyone has a certificate, and in the mind of whoever signed the cheque the skills problem is closed.
Then I look at what happens three months later. Some people use AI every day and have changed the way they work. The rest opened ChatGPT twice, got distracted, and nothing about their working day has changed. The course certified attendance, not learning. And nobody in the company could tell you today who is in the first group and who is in the second.
The black hole I have seen before
This problem, "we do not know who actually learned", I have seen identically in a completely different context. Working on AG Academy, a high-ticket training product for companies of 20 to 60 people, the problem was not the quality of the course content. It was that after the sale, students vanished into what we internally called the black hole: nobody knew who had received their access, who had been added to the group, who had actually started the programme. A quarter of students asked for a refund in the first 14 days, and when we went to understand why, the recurring answer was that they had felt lost, unsupported, unsure where they were in the programme.
The fix was not one more course for the team handling onboarding. It was a system that made every single student's progress visible in real time: who had received what, who had started, who had stopped and where.
The parallel with internal AI training is direct, even though the context is different: there it was clients in a delivery programme, here it is employees in an adoption programme. But the structural flaw is the same. A one-off course for the whole team has the same black hole as onboarding without tracking: nobody knows who genuinely learned to use AI in their daily work and who just sat in the room.
Why the real brake is not what you think
The numbers confirm the problem is real, not an impression. When Istat asks companies that evaluated AI but did not adopt it what the main obstacle was, the most cited answer is not the cost of the tool nor the maturity of the technology: it is insufficient skills (Istat, Enterprises and ICT, 2025). It is a figure that inverts the priority most companies bring to AI: they start by looking for the right tool, when the thing actually blocking adoption is upstream, and it is human.
Buying the tool is the easy part. Making it part of somebody's daily work is the hard part, and that is where a one-off course, however well delivered, stops.
Cost is not the real block. Skills are.
What works: a reference point per department, not an event
The alternative we see working is targeted rather than sprayed across everyone: an AI Champion per department, not necessarily a technical profile. A person who:
- experiments with tools on their team's real processes, not on generic exercises,
- works out what does and does not work in that specific context,
- turns what they discover into a written procedure,
- and passes it to colleagues, staying the reference point when something jams.
Experiments on the process
Tests tools and prompts on their department's real cases, not on classroom exercises.
Codifies into procedure
Turns what works into written, versioned instructions the others can consult.
Spreads and holds the line
Passes the practice to colleagues and stays the reference point when something jams.
This model has already led to training more than 2,500 people on AI applied to real work, not theory. The difference from a one-off course is that here there is always somebody who can answer the question "who is actually using it": the AI Champion themselves, because they are there every day watching. You do not need to build a separate tracking system: the human presence does that work by nature. It is also how knowledge accumulates instead of dispersing, especially when it sits on a shared Company Brain, where the procedure written by one department stays available to all the others.
The skills that really count
Good news for anyone worried about having to hire engineers: the skills that make the difference for most roles are not technical, they are practical.
- Knowing which tasks to delegate. Understanding which pieces of your work make sense to give AI and which do not, instead of trying to delegate everything or nothing.
- Writing clear instructions. A good prompt is not a magic trick, it is clarity: context, goal, explicit constraints.
- Assessing the output critically. Recognising when an answer is solid and when it has to be verified line by line. Hallucinations exist, and somebody who cannot recognise them is at more risk with AI than without it.
- Integrating into the existing workflow. Putting AI inside a process that already works, rather than running it alongside like a toy disconnected from the real job.
Coding is relevant to a few specialist roles. The rest is applied literacy: the same thing the AI Champion practises every day and translates into procedure for the others.
| Skill | What it means in real work | Needed by |
|---|---|---|
| Delegating the right tasks | Choosing what to give AI and what to keep, instead of trying all or nothing. | all roles |
| Writing clear instructions | Context, goal, explicit constraints. No tricks: it is clarity. | all roles |
| Assessing the output | Recognising when an answer is solid and when to verify it line by line. | all roles |
| Integrating into the flow | Putting AI inside a working process, not running it alongside as a toy. | process owners |
| Coding | Only relevant to whoever builds integrations or custom agents. | technical roles |
"The right question is not 'have we trained everyone', it is 'can we name the people who use AI every day in their own work'."
It is not just common sense: it is also the law
Since 2 February 2025, article 4 of the European AI Act requires anyone developing or using AI systems to ensure an adequate level of literacy among the people working with them, smaller companies included. Having people competent on AI, verifiably so, is no longer just a competitive advantage: it is a regulatory requirement. We cover it in more depth in the article on the AI Act and AI literacy.
How to start
- Pick a department where a quick return is plausible and measurable.
- Identify the right person as AI Champion: knowledge of the process comes before curiosity about the technology.
- Have them work on a real case from the department, not on theoretical exercises.
- Turn what works into a written, shared procedure.
- Extend to other departments, replicating the model instead of repeating the course.
The right question to ask
It is not "have we trained everyone", it is "can we name the people who use AI every day in their own work". If the answer does not come immediately, the problem is not the technology you bought: it is that you do not yet have somebody keeping it alive, department by department.
Want to build internal competence that stays? Have a look at the AI Champion programme. If instead you want your team to start from the basics, send whoever you like to Learn AI.