There's a ritual now, preceding every major tech conference of the year: someone on stage utters the phrase 'this will change everything,' and half the room applauds while the other half mentally calculates how many months their job will be obsolete. Meanwhile, in the more discreet layers of the real economy, hundreds of thousands of jobs disappear, silently, without press conferences or standing ovations.

We are told that this shift is nothing new, that every technological revolution has destroyed jobs before creating others: the automobile killed the blacksmith but gave birth to the mechanic, computing eliminated the typist while birthing a colossal software industry. This is true. It is also a convenient way not to look at what is actually happening: not a replacement of tasks, but a change in the very nature of intellectual work—and a shift in who, exactly, benefits from it.

Destruction Is Faster Than Creation, And That's No Coincidence

Let's start with the numbers, as that is the language this debate demands to be taken seriously. Amazon has eliminated nearly 30,000 positions in recent months, a move followed by Google, Microsoft, Meta, and Intel—at the very moment when the cumulative valuation of AI labs reached heights no serious analyst would have dared predict three years earlier.

But the most telling figure is not the number of cuts: it is the reasons cited. A *Harvard Business Review* survey published in January 2026, conducted among over a thousand executives interviewed in December 2025, shows that 39% of them admit to having reduced their workforce in anticipation of future AI capabilities—compared to only 2% based on actually observed automation gains. Translation: most current cuts do not respond to an effective replacement of human work, but to a bet on this replacement, made at the expense of employees laid off even before the technology has proven itself.

We can elaborate on studies that promise the emergence of new professions—data architect, AI ethicist, algorithmic content curator. These professions exist, or will exist. The report by the General Directorate of the Treasury published in June 2026 acknowledges this unequivocally, citing the work of economist David Autor: 60% of workers today hold jobs that did not exist in 1940. Economic history proves the optimists right in the long run.

The problem is not the existence of these professions, it is their relative scarcity compared to the volume of positions they claim to offset in the short term. The McKinsey Global Institute speaks of 30% of working hours being automatable by 2030 in advanced economies; the OECD puts the share of French jobs exposed to a high risk of automation at 27%, or more than four million positions—concentrated in administrative support, accounting data entry, and translation. The World Economic Forum, meanwhile, forecasts a net positive balance of 78 million jobs created globally by 2030. But an automated task does not automatically equate to a suppressed position—except that, in practice, that's exactly what most general managements do as soon as margins allow. The time saved is not redistributed towards more creativity; the workforce is reduced, and the gain is distributed to shareholders.

The Myth of the Augmented Worker

The dominant discourse is that of the 'augmented worker': AI as a co-pilot, as a cognitive prosthesis that frees humans from mundane tasks to refocus them on judgment, creativity, and relationships. It's appealing, and not entirely false—I've seen colleagues save valuable time on synthesis tasks they hated. But a crucial question remains dodged: whose is this liberated time?

In a company that treats its employees as a cost center, the time freed up by AI is never returned in the form of fewer working hours or greater autonomy. It is immediately reinvested in more production, more demands—until the workload reaches its previous saturation level, but with a reduced staff. This is called a productivity gain. It should be called, more honestly, an intensification of work disguised as progress.

What We Lose When We Lose Boredom

There's an aspect of this transformation that is rarely discussed: the disappearance of what I would call productive boredom. Those moments of repetitive tasks, laborious writing, tedious research, which seemed to have no value in themselves but which shaped a skill, an intuition, a memory of the craft. The young journalist who spent hours sifting through archives to find a figure developed a detailed knowledge of their subject that instant access to a synthesized answer will never give them. The intern lawyer who manually drafted dozens of similar conclusions eventually internalized a grammar of law that automatic text generation simply bypasses.

One might argue that this is the reasoning of every generation facing every new technique—scribes deplored the disappearance of oral memorization with the advent of writing. The argument is valid. But there is a difference in degree, if not in nature, between delegating calculation to a calculator and delegating reasoning itself to a system that produces a plausible result without the user needing to understand how it got there. The risk is not only the loss of a skill: it is the loss of the ability to evaluate whether the result produced by the machine is correct, for never having done the exercise oneself.

Jobs That Resist, And Why That's Not Reassuring

Data from France Travail confirms the quiet trend: an explosion of offers for tech-AI profiles, paralleled by a collapse in recruitment for traditional administrative jobs. To compensate, we are told that relational, manual, or real-world jobs—caregivers, artisans, educators, plumbers—are safe. This is probably true in the short term, but it is small consolation.

Firstly, these jobs are precisely those our societies have remunerated the least, even though they should, in the hierarchy drawn by cognitive automation, become the rarest and thus the best paid. Nothing indicates that this shift is underway: a prompt engineering engineer continues to be paid more than a nursing assistant, even though the former profession might disappear within five years and the latter will never disappear.

Secondly—and this is the most uncomfortable point—the resistance of these jobs to automation is not necessarily good news for those who hold them. It signifies a society polarized between a cognitive elite orchestrating machines and a mass of physical presence workers, deemed low-skilled by the market: two parallel economies that intersect less and less.

Moving Beyond Fatalism Without Falling Into Denial

I do not believe we should succumb to catastrophism, nor advocate for technological degrowth that has no chance of occurring in a world of international competition. The question is not whether AI will continue to transform work—it will. The question is who decides the distribution of the gains that this transformation generates, and according to what criteria.

This requires concrete political choices: taxation that captures a share of productivity gains related to automation to fund the retraining of displaced workers; a sincere revaluation of human presence jobs; education that stops preparing young generations for jobs that will no longer exist, to orient them towards what machines will structurally never do well—uncertain judgment, assumed responsibility, embodied relationship. It also requires, more modestly, that everyone ask themselves the uncomfortable question: in my own work, am I delegating boring tasks to the machine, or am I delegating my ability to understand what I produce?

What Remains for Us, Against All Odds

There's something ironic about writing this text at a time when the information industry itself is reinventing itself with artificial intelligence, where one can read, published on the same day, an article praising a model's performance on an agency benchmark and another announcing tens of thousands of job cuts in the press. This tension is not resolved, and it probably never will be fully. But if there's one thing that neither the most sophisticated models nor the most skillfully presented redundancy plans can take away from us, it's the responsibility to collectively choose what we want to do with this liberated time. Entrusting it solely to quarterly profitability logics would be the biggest mistake of this decade.

Editorial Opinion

This opinion piece addresses a blind spot in the debate: we endlessly discuss the number of jobs destroyed, almost never what automation does to learning a trade. The OrChair editorial board particularly highlights the *Harvard Business Review* figure—39% of executives laying off employees in anticipation versus 2% based on actual gains. It says the essential: a large part of the current social cost of AI is not technological; it is decisional. And a decision, unlike a law of physics, can be made differently.

*Opinion. The views expressed are solely those of the author.*