Reading note. This note starts from a Clubic article ("no, artificial intelligence will not destroy your job, unless…") and traces back to the studies it relays. We checked the sources: the article relies on the International Labour Organization report, not on a comparison with the personal computer. That historical analogy is ours, and we own it as a reading frame, not as proof. The figures cited come from identifiable, dated publications, linked at the bottom of the page.

In one sentence

Contrary to catastrophist predictions, employment statistics show — for now — no massive wave of job destruction. But that apparent calm hides a quieter shift: according to the International Labour Organization (ILO), one job in four is exposed to generative AI — most often transformed, not eliminated. And where the number turns worrying is among juniors: a Stanford study already measures a decline in hiring for 22-to-25-year-olds in the most exposed occupations. The danger is not the disappearance of jobs. It is the closing of the entry doors.

1. What the study everyone cites actually says

The Clubic article going around rests on a precise source, and it is worth naming it correctly. This is not a consulting-firm prophecy, but work by the ILO, a United Nations agency, carried out with Poland's NASK institute and published in May 2025 under the title Generative AI and Jobs: A Refined Global Index of Occupational Exposure.

The report makes three findings that we take as they stand, because they are methodologically solid:

  • 25% of global employment falls within occupations exposed to generative AI — and up to 34% in high-income countries.
  • The most likely outcome is not replacement but transformation: most jobs are made of tasks, only some of which can be automated.
  • Exposure is higher for women, a larger share of whose employment falls into the most exposed categories.

In other words, the reassuring headline hides a less reassuring nuance. "No, AI will not destroy your job" ends with "…unless." And the "unless" is the whole point.

2. History repeats — which is actually reassuring

When the first computers arrived in companies, people predicted the end of secretaries, accountants, administrative staff. When the Internet arrived, retail, the press and travel agencies were declared dead.

The economy did not stop. It transformed. Some jobs vanished, others appeared, and productivity eventually rose.

Economists have a name for this lag: the Solow paradox — the idea that you can see computers everywhere except in the productivity statistics. The gain does not follow adoption immediately; it arrives with a delay, the time it takes for organisations to learn to reorganise around the tool. That is exactly the phase we are in with AI: much noise, still little trace in the macroeconomic employment figures.

"We see AI everywhere except — for now — in the job-destruction statistics. The historical precedent calls for patience, not for blindness."

That precedent is reassuring on one count and worrying on another. Reassuring, because it suggests the employment apocalypse will probably not happen. Worrying, because in every wave, the ones who suffered most were not the jobs, but the people who refused — or were unable — to adapt.

3. The right question isn't "how many jobs?"

The debate is badly framed. People keep asking: "how many jobs will disappear?" The useful question lies elsewhere.

AI does not replace humans. It replaces tasks — repetitive, predictable, documented ones. That is precisely what the ILO measures by reasoning task by task rather than job by job. And those tasks used to fill a large part of our days:

  • writing minutes and reports;
  • searching for and synthesising information;
  • fixing or producing simple code;
  • answering routine questions;
  • formatting presentations.

The consequence is mechanical: a worker assisted by AI produces more. So the real question is not "how many jobs will AI eliminate?" but "how many people will refuse to use it?" It is not technology that replaces humans. It is the humans who use it better who replace those who do without.

4. The end of the developer-king

For twenty years, the digital industry placed the developer at the centre of everything. The company had an idea, the developer built it, and development time was the scarce resource.

That scarcity is fading. Code-generation tools improve month after month, and the ability to write code is no longer the limiting factor. What becomes limiting again is what should always have been: understanding the problem.

We wrote this already in Software Engineering Is Not Dead — it is not the end of developers, it is the end of their monopoly on value. And in When the Token Costs Almost Nothing, we showed the same curve for AI itself: when a resource becomes abundant, it stops being an advantage. The return of the project lead, the designer, the domain expert, the strategist is not nostalgia. It is the logical consequence of a world where knowing where to go is worth more than knowing how to get there.

5. The real risk: the collapse of the learning ladder

Here is the most worrying signal, and it is now quantified.

Traditionally, a professional progressed by steps: junior, confirmed, senior, expert. Yet AI automates exactly the tasks that served as the entry rung — research, drafting, simple development, documentation, first diagnostics. Those are the tasks on which a junior learned the trade.

The Stanford Digital Economy Lab brought this into focus in an August 2025 study with a telling title: Canaries in the Coal Mine? Drawing on payroll data from ADP, the largest US payroll provider, the researchers find that since the mass adoption of generative AI, employment among 22-to-25-year-olds has fallen by roughly 13% in the most exposed occupations — while employment for more experienced workers, in the same jobs, stayed stable or grew.

If this signal holds — and it is still a recent measure, in a single country — the risk is not a mass disappearance of jobs. It is the silent disappearance of the entry doors. A company that can have a model produce junior-level work will look for immediately operational profiles. But if no one hires juniors anymore, where will tomorrow's seniors come from? We are sawing off the branch the next decade's experts sit on.

6. Who will win

The winners of this shift will not necessarily be the best technicians. They will be those able to understand a trade, frame a problem, orchestrate several tools, decide, communicate — in short, to create value rather than produce volume.

AI does not replace vision: it amplifies it. It does not replace strategy: it accelerates it. It does not replace human intelligence: it augments those who know how to use it. This is the same conclusion as for code and for the token. The raw material becomes abundant; scarcity moves toward discernment.

7. A situated word

From Réunion Island, 9,000 km from Silicon Valley, this story reads with a mix of optimism and vigilance.

Optimism, because a technology whose cost collapses always ends up reaching the peripheries. AI is leaving the giants for SMEs (small and medium-sized enterprises), then individuals, then territories. For a small association, a craftsperson, a town hall, it is a real chance to do, with three people, what once required a large team.

Vigilance, because the closing of the entry doors hits hardest where they were already narrow. Where juniors are trained with difficulty, watching the learning ladder fold up would be bad news. Our frugal-lab bet stays the same: not to wager on the quantity of AI consumed, but on the relevance of its use — and to keep, stubbornly, training humans able to ask the right questions.

Because the arrival of the computer did not destroy work. Neither did the Internet. AI probably won't either. But it is already destroying one thing: the excuse that we still have time to wait.

Sources and further reading

This document is updated if new elements appear. Last revised: 文 June 21, 2026.