VivaTech fascinates, the office stagnates. 70% of employees learn AI on their own. The value of AI leaks to individuals, not businesses.
Last week, leaders flocked to VivaTech. They returned with their heads full of innovations: autonomous agents, multimodal models, intelligent co-pilots…with the certainty that their sector is being transformed.
They are right. And completely wrong at the same time.
Because while the demos took place on the stands, a stubborn reality waited in the office. According to the Master The Monster-Kantar study, 70% of employees who use AI learned it on their own. Only 9% were trained by their company. 9 out of 10 workers also believe that their company does not have a clear AI strategy and for 1 in 2, their company is behind in terms of tech and AI maturity.
What good are dizzying AI innovations if no one in the company is able to use them other than as an improved search engine? Companies seem to be abandoning training and, more profoundly, seem to be hesitant in the face of AI.
When organizations don’t support their employees, they don’t stop. They tinker. In France, 46% of workers use their personal AI tools for professional tasks. This Shadow AI is often presented as a security risk. This is true of course, but it misses the point.
Shadow AI is not a rebellion. It’s a rational response to a void. Since the company offers nothing, or nothing suitable. In fact, only 25% of workers believe that the tools provided meet their needs. Employees therefore make do with what they have. The main reasons: speed (23%), ahead of simplicity (16%) and efficiency (14%). No desire to bypass IT. Just the desire to work.
It is illusory to believe that this situation is neutral for the organization: if employees save time, the company hardly benefits.
A study published in April 2026 by researchers from Stanford and Carnegie Mellon demonstrates this: the value created by generative AI for its American users reaches 172 billion dollars per year, far ahead of the revenue that AI players earn from it (14 billion). The value of AI is captured by individuals, not organizations.
When an employee gains two hours per week thanks to ChatGPT without their employer measuring or organizing it, these two hours add nothing. They enrich the individual and incidentally the shareholders of OpenAI. Not the company. This is the silent hemorrhage: a leak of value that no one has decided to stop, because no one has yet decided to take the subject head on.
In the United States, the training problem also exists: 42% of American employees say that their employer expects them to learn AI on their own. But where France differs profoundly is in the involvement of management. Across the Atlantic, it is senior executives who do the most Shadow AI, 65% of them exactly, compared to 31% of employees. American management seized the tool from the top, even in a disorderly manner. In France, it’s the bottom of the pyramid that tinkers while the top talks about transformation.
America has an alignment problem. France has a commitment problem. It’s not the same depth of delay. And the AI market – the models, the platforms, the data – remains overwhelmingly American. If French organizations do not take up the subject, they will finance the value creation of others.
This debate goes beyond the question of current productivity. It involves the very form of tomorrow’s work.
“Today, our engineers no longer write a line of code. (…) You ask agents to write the code for you. You give specifications, you give orders. It’s a profound change.”
Arthur Mensch, CEO of Mistral AI, hearing at the National Assembly, 2026.
If agent orchestration is the next labor market norm – and Arthur Mensch doesn’t talk about it lightly – then the question of training is no longer just about today’s productivity.
It concerns the ability of French companies to survive in a world where professions are being rewritten. Moving from execution to orchestration, from doing to directing agents, is not a marginal increase in skills. It’s a reinvention of the entire organization.
This reinvention will not happen without a strategy, that is to say without answers to the questions that AI poses to companies: the role of middle management, the place of young people, the organization of decision-making autonomy, the border between transparency and surveillance, the redefinition of the very notion of competence…
But time is running out and AI is moving fast, very fast. The transformation will have to be much faster than what companies have experienced in digital. This is the sine qua non condition for capturing the value created and not suddenly losing ground compared to American companies.
