AI jobs in France may be giving one of the clearest early signs of how artificial intelligence could alter careers without producing a sudden wave of redundancies. New analysis from the Banque de France finds that permanent recruitment has fallen more sharply for young adults working in occupations whose tasks are especially exposed to generative AI. The important distinction is between people losing jobs and people struggling to get their first one. The central bank’s research identifies an association, not proof that AI caused the hiring gap. That caveat matters when young workers and employers are trying to interpret an already difficult labour market.
The Banque de France analysis, published on 8 October 2026, compared French employment movements through March 2026 with an occupation-level measure of AI exposure. Among people aged 21 to 29, permanent hires in the most AI-exposed occupational group were about 28 per cent below their November 2022 level by March 2026. The comparable fall in the least exposed group was about eight per cent. These are changes relative to earlier hiring levels, not a claim that 28 per cent of young people have become unemployed. Nor is it a clean experiment in which some firms randomly receive AI and others do not.
What the AI jobs in France figures actually measure
The research uses administrative employment-movement records rather than asking workers to guess how many jobs might disappear. It distinguishes new permanent-contract hires from departures, separates age groups and compares categories of work according to the tasks that software could plausibly assist or automate. That produces a different picture from surveys asking people whether they fear replacement. It makes the recruitment channel visible: companies can change the number and type of vacancies they offer while the existing workforce remains in place. In a country with substantial protections around permanent employment, that channel deserves particular attention.
The central bank describes a roughly 20-percentage-point gap between the hiring trajectories of the most exposed and least exposed occupations among workers aged 21 to 29. It says the gap narrows for people in their thirties and forties, and almost disappears among workers aged 50 and over. The study also tracked permanent-contract terminations, which declined among younger workers in highly exposed roles. Hiring fell more sharply than departures. That combination points towards slower movement into and between positions rather than a dramatic increase in dismissals. A quiet change at the entrance to employment can be difficult to see in the national headline figures.
Researchers constructed their exposure measure from the skills listed in France Travail’s occupational directory. A language model assessed which tasks AI might perform or assist with, while accounting for activities requiring physical presence, tacit knowledge or human interaction. This was a measure of technical potential, not proof that each company actually automated those tasks. The team therefore cross-checked against information on reported business use of AI. That helps test whether exposure has practical meaning, but still cannot establish what caused an employer’s particular decision not to replace a departing junior colleague.
The firm-level comparison covered around 6,000 companies, including nearly 600 reporting intensive AI use at the beginning of 2026. Permanent hiring among young people had followed a less favourable path in the intensive-use group, while total employment did not show a statistically clear divergence. Crucially, adoption was not random. Firms that became heavy AI users may already have differed in the work they did, their hiring plans and their response to other pressures. A correlation that survives several checks is useful evidence. It is still not permission to say a chatbot eliminated a particular vacancy.
Why the first career step can change before jobs disappear
Imagine a company whose experienced employees can now produce an initial report, marketing draft or code review more quickly. It may not need to dismiss anybody. Instead, when a junior employee leaves, the business might redistribute the workload among senior staff using new tools. If the next graduate vacancy is never advertised, the adjustment is almost invisible to customers and existing colleagues. The person who loses out is someone the organisation has not met yet. This is a possible mechanism, not a recorded account of what the surveyed French companies actually did.
There is a longer-term question hidden inside that decision. Junior employees often learn by doing work that more experienced colleagues can check. If technology absorbs every straightforward assignment, a business has to deliberately create new routes for beginners to acquire judgement. It might move training towards verifying AI output, investigating unusual cases, interviewing clients and understanding why a final decision was made. Those activities can produce better skilled workers, but they require time and supervision. Simply expecting new employees to arrive with senior-level judgement risks creating a shortage of expertise later.
LiveAIWire has previously covered research in which AI-adopting companies expanded senior roles faster than junior ones. That earlier analysis examined a different international dataset and found changes in the composition of employment, rather than a statistically clear collapse in the number of junior staff. The French study sharpens a related question by following permanent hiring directly and comparing younger and older age groups. It should not be described as a replication of every finding from that earlier work. What the separate results share is a reason to examine who is entering firms, not merely how many employees those firms have.
The distinction also matters when someone reads job advertisements. A vacancy requires a budget, a suitable applicant and a hiring decision; the number of advertised roles is not the same as the number of people employed. Employment data, by contrast, tells us whether people actually entered or left positions, but not why a manager felt confident doing without an additional hire. Combining measures is more useful than treating any single figure as an answer to the entire employment question. The broader Stanford AI Index account of the economy likewise stresses that productivity findings are emerging while workforce effects remain uneven and difficult to separate from other changes.
Why the central bank does not call this an AI verdict
The period since late 2022 includes higher borrowing costs, a correction after extraordinary technology-sector hiring and major changes in remote working. Those developments can strike occupations that are also attractive candidates for AI assistance. A company that later reports heavy AI use may have been cutting recruitment for unrelated reasons before the latest models appeared. Banque de France explicitly says its observational comparison cannot separate AI’s precise effect from the characteristics of the businesses choosing to adopt it. Reporting the result as a proven automation-driven employment shock would therefore go beyond the evidence.
The cross-country record is mixed as well. The French authors refer to US findings consistent with pressures on younger employees, but also to research in Norway and Denmark that does not show the same average employment or earnings effects. Labour protections, industrial structure, measurement and the timing of adoption all differ. This is why a result in France does not provide a direct forecast for Britain, the United States or every graduate looking for work. It does suggest an indicator those countries can measure for themselves: recruitment by age and task exposure, compared with departures and overall staffing.
Another complication is the difference between tasks and occupations. A job is a collection of responsibilities. A model that assists with standard drafting cannot necessarily replace the person who understands a customer’s problem, takes responsibility for an error or notices that a situation falls outside the usual pattern. The sensible question for a worker is not whether an occupation appears on a generic danger list, but which parts of the work may change and which abilities will remain necessary to judge the output. Employers face a corresponding choice about whether to use saved time for deeper work or simply lower the hiring budget.
What young applicants and employers can take from it
An applicant cannot fix the national recruitment market alone. But the research strengthens the practical case for building evidence of judgement alongside evidence of software familiarity. Reviewing an automated answer against original material, explaining a correction and showing a completed real-world project demonstrate more than the ability to type a prompt. Applicants can also seek roles where supervision and progressively harder assignments are explicit. These are reasonable responses to uncertainty, not guarantees that an individual application will succeed.
For employers, the more difficult discipline is tracking who is being trained. A firm can congratulate itself on getting more work done with fewer junior hires while unknowingly weakening its future management pipeline. LiveAIWire has also reported how AI adoption can reshape the balance between junior and senior recruiting across industries, and why presenting AI as a route to more meaningful work can change employee engagement. Those stories point to choices managers still control. Productivity improvements and credible early-career pathways need not be enemies if training is treated as an investment rather than spare work for someone to perform.
The most telling future evidence will come from following young cohorts over time, separating workers who move quickly into responsible positions from those whose first opportunities never materialise. Researchers will also need stronger ways to identify when and how businesses actually deploy AI. For now, the central bank has not announced the end of the graduate job. It has found a measurable hiring divergence at the bottom of the career ladder, precisely where a small change today can have an unusually long shadow over someone’s working life.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.