Imagine a ladder whose first rungs had been sawn off. Those already at the top keep climbing. Those waiting at the bottom no longer know how to get up.
That, in one image, is what a study published this week by the Banque de France and the HEC laboratory devoted to the future of work describes. It covers more than 1,500 occupations, and its conclusion is clear: in France, artificial intelligence closes the door on beginners first.
The figures
The researchers compared the occupations most exposed to generative AI, those where it can do part of the work, with the least exposed occupations. They tracked hires from November 2022, when ChatGPT arrived, through to March 2026.
For 21-29 year-olds, permanent-contract hires fell by 28.3% in the most exposed occupations, against 7.5% in the least exposed. Annualised, that represents roughly 75,000 fewer permanent contracts for this age group.
For young developers and IT engineers, the drop reaches 42%.
Across all contract types, hires of young people fell by 22.1% in exposed occupations, while they rose slightly, by 1%, in more manual occupations.
And the gap depends heavily on age: it reaches nearly 21 points among 21-29 year-olds, against less than 4 points among the over-50s.
Why young people first
The reason lies in the nature of an occupation's entry-level tasks. A first draft of a report, a summary, a spreadsheet, a simple piece of code: these are precisely what used to be handed to beginners so they could learn. And they are precisely what an AI can now do.
Experienced employees, meanwhile, are protected by what AI does not yet replace: their knowledge of the job, their judgement, their network. The study says as much in its own way: the older you are, the more the gap between exposed and non-exposed occupations narrows.
Another telling sign: in exposed occupations, young people on permanent contracts quit far less, 23% fewer departures against 6% elsewhere. When the market closes, you cling to your job. And fewer departures means even fewer openings for those coming in.
Hiring depends on many things: the economic climate, interest rates, the health of a sector. The strength of the study is that it compares, over the same period and in the same economy, occupations highly exposed and barely exposed to AI. Both groups went through the same economic conditions. What separates them is their exposure to AI. That does not prove that AI explains everything, particularly in IT, which is affected by other difficulties. But a gap of more than 20 points, concentrated among the youngest, is hard to put down to chance.
What it means
This study complements the one from the Paris School of Economics we discussed in late September, which showed that recruiters were demanding fewer and fewer IT skills. Both tell the same story of transformation from two angles: AI does not necessarily eliminate occupations, but it changes who gets hired, and to do what.
The risk is a problem that will only become visible later. Tomorrow's experts are today's beginners. If no one hires juniors any more, who will know, in ten years, how to do the work that AI does not do?
If you are just starting out
Nothing is settled, but some things matter more than they used to. Real experience, on a work-study placement or internship, is becoming a decisive asset. Knowing how to use AI is not enough: you have to be able to judge what it produces, and spot when it gets things wrong. And jobs that combine technical skill with human contact, or technical skill with fieldwork, hold up better.
What we take away
It was assumed that AI would affect the most repetitive roles first. This study shows that it affects the youngest first, those who were starting their careers with those tasks.
If you saw through the first rungs of the ladder, other ways of climbing will have to be invented. This may be the most urgent task that AI imposes on companies and schools alike.