In my first piece, I wondered about the gap between the speed of machines and the pace of human beings. About the difficulty of welcoming progress when it touches our work, our place in the world, sometimes our sense of who we are.
Since then, another question has been on my mind: what are we doing with the time this adjustment takes?
A transformation can be slow to become visible while its consequences move very quickly. The office is still there. The job has the same name. Meetings happen at the same times. Yet hiring may slow, certain skills may become less scarce, and value may start moving elsewhere.
We could grow accustomed to this apparent continuity just as it becomes misleading.
I do not know whether, in five years, machines will be able to perform most of our intellectual tasks. Still less when robots will be able to operate everywhere we work today. Pulling off a demonstration and providing a reliable service, at the right price, under unexpected conditions, remain two different things.
But machines do not have to know how to do everything for a great deal to change.
A company can rethink its staffing without fully automating a profession. A profession can remain indispensable while employing fewer people. And a new graduate may discover that the tasks through which they were meant to learn their trade are precisely the ones being automated first.
The ILO distinguishes between the exposure of tasks, the transformation of jobs and their disappearance. Its 2025 study considers transformation more likely than wholesale replacement. That is a useful distinction, not a personal guarantee of peace of mind. [1]
What worries me is the possibility of a very uneven transition. Some will have time to learn, invest and move their activity elsewhere. Others will discover the change when they already need to respond to it…
Adoption will take time, of course.
A tool arrives in an organisation that already has its habits, commitments, rivalries and a few procedures whose origins nobody quite remembers. The actual work has to be understood, results checked, responsibilities assigned. Some of this slowness is necessary.
Some of it is less so.
A manager may have an interest in keeping a large team. A contractor may earn more by maintaining a complication than by solving it. A department may call for an ambitious transformation while asking that it change neither its budget nor its powers. A complete revolution, subject to keeping the organisation chart…
But it would be too easy to reserve this criticism for established institutions. The newcomers have interests of their own: selling fast, making customers dependent, getting others to finance infrastructure or bear the consequences of failure.
The word “innovation” does no more to excuse us from examining incentives than the word “tradition”.
That is also why I do not want to lump every form of resistance together. Someone protecting a privileged position and someone asking what will happen to their income are not defending the same thing. An objection about reliability, responsibility or the sharing of gains deserves more than a sigh about France falling behind.
If we ask someone to help transform their work, we need to be able to talk to them about what comes next. An abstract promise of prosperity does not pay for the intervening years.
This transformation also has a physical reality that demonstrations make easy to forget.
Behind the software, there must be electricity, networks, chips and cooling equipment. Behind the robots, there are parts to manufacture, maintain and replace. All of this calls for land, workshops, skills and time.
The International Energy Agency already points to constraints on grid connections and supplies of electrical equipment. It also highlights the opposite risk: building for overestimated demand, based on projects that will not all materialise. Capacity needs to be prepared without treating every investment announcement as a firm order. [2]
This is where the question becomes political.
For France and Europe, having a few appealing applications will not be enough to guarantee the ability to act. We will need to know what we control, what we buy and what we might lack.
I do not believe in a magic shopping list that amounts to making everything ourselves. It is easy to write, much harder to fund. We will have to choose industrial strengths to build on, develop skills, organise cooperation and keep alternatives where dependency would become dangerous.
Energy is part of this. For France, that means, among other things, the operating life and renewal of the nuclear fleet, but also grids, connections and the people capable of building them. Expertise in components, computing and certain industrial applications matters too.
Owning equipment is not enough, either. We need to know how to operate it, repair it and keep going when its supplier becomes too expensive, unavailable or politically out of reach.
Our institutions will also have to change. On this point, impatience seems justified to me — provided we spell out what we want to accelerate.
We should be able to try a tool quickly when the cost of failure is limited and reversible. We should also be able to stop an experiment that fails to deliver, without having to rescue the presentation that announced it.
Conversely, when an error affects a patient, a right or essential infrastructure, the demand for evidence must increase. The European “high-risk” classification imposes obligations; it does not mean a blanket ban on using AI in the fields concerned. [3]
The task is therefore more precise than “removing constraints”. It means distinguishing those that actually protect people from those that complicate the work without protecting very much. Then giving teams the means to try, measure and correct.
The same applies to education. Learning to use a tool must not remove the learning that lets us notice when it is wrong. We will need to look closely at the steps we save time by skipping: some may have been unnecessary; others were training our judgement.
The hardest part remains: helping people make it through this period.
We readily talk about retraining as though it were a fairly simple move from a “declining occupations” column to a “jobs of the future” column. Between the two, though, there is an age, a home, children, an income to maintain and sometimes several years of training.
The emergence of new jobs does not automatically solve the situation of those who lose theirs. They still need to be able to get those jobs, where they live, at an acceptable wage.
We will need to fund time to learn, build credible routes into new work and discuss how gains are shared before decisions are made. Asking for trust after leaving everyone to guess what they might lose strikes me as a rather poor method.
I remain impatient to see what these tools will make possible. But I would like our enthusiasm to make room for a few very ordinary questions: who will know how to make all of this work? Who will have a choice? Who will pay when we get it wrong?
And we will need time to build the answers.
It would be a shame to spend that time explaining to ourselves that we still have some left…