French firms using AI heavily hire up to 25% fewer young people on permanent contracts

"A company that entrusts AI with some of the tasks usually performed by beginners does not necessarily need to make redundancies. It can simply recruit fewer people and not replace some departures." This is the possible explanation put forward by Antonin Bergeaud, associate professor at HEC Paris, in the article published by the school on 9 October. His team's figures describe this pattern. Among intensive AI users, hiring of young people on permanent contracts is falling. Their total headcount does not differ significantly from that of other companies.

The study is signed by Antonin Bergeaud and Oscar Degos (HEC Paris, LIFT), Marco Guido Palladino and Antoine Bertheau (Banque de France) and Simon Bunel (INSEAD). The LIFT note 2026-02, "Generative AI: a hiring freeze for young people in exposed occupations", presents the results, the working paper the method. The Banque de France bulletin no. 463, posted online on 8 October and updated on the 13th, summarises them without quantifying the 17 to 25% gap. The work "reflects the opinions of the authors and not necessarily those of the Banque de France", the note states.

600 intensive users compared with other companies in their sector

The authors cross-reference two sources. Employers' social security declarations in the private sector, processed by Dares (the French ministry of labour's statistics directorate), provide labour flows from January 2019 to March 2026. Around 6,000 companies surveyed by the Banque de France's monthly business survey are matched. Around 600 report intensive use of AI. Quarter by quarter, the model compares them with other companies in their sector, including those that use AI in a limited or experimental way.

"After the fourth quarter of 2022, permanent-contract hiring of young people falls sharply among intensive AI users," the authors write. Depending on the quarter in 2025, these companies hire 17 to 25% fewer 21-29 year-olds on permanent contracts than others, a gap of -0.19 to -0.29 log points. Among older employees, gaps remain small and rarely significant. Total headcount, meanwhile, "does not diverge in a statistically significant way" from that of other companies. Estimates drift towards -2 to -3% at the end of 2025, without being distinguishable from zero.

The Banque de France bulletin presents this result as an association between reported adoption and changes in hiring. Usage is reported in early 2026, while gaps appear as early as 2023. "We cannot assert that the intensive users of 2026 were already using AI" when their hiring began to fall, the authors write. They do not rule out a selection effect: companies that were already reorganising their recruitment may have adopted AI later.

Within the same company, exposed occupations hired a third less

The note also compares occupations with each other, within 541 intensive users matched to Dares data. In the first half of 2025, the most exposed occupations hire about a third less than other occupations in the same company. The comparison is made within each company. According to the authors, the gap therefore cannot be attributed to a demand shock specific to the company. It narrows at the end of the period, when it is no longer significant.

75,000 permanent contracts: missing hires over a year, not destroyed jobs

The figure of 75,000 permanent contracts, headlined by several media outlets, comes from another analysis in the note. It does not rely on reported usage. It classifies 310 occupations into five quintiles of exposure to generative AI. Among 21-29 year-olds, permanent-contract hiring in the most exposed quintile is 28.3% below its November 2022 level in March 2026, compared with 7.5% in the least exposed. Relative to the trend level of November 2022, around 22,000 hires per month, the decline reaches about 6,200 permanent contracts per month. "Over a year, this represents around 75,000 fewer permanent contracts for 21-29 year-olds alone," the note writes.

This total counts hires that did not take place, compared with the 2022 level. It includes the decline common to all occupations. Only 4,600 of the 6,200 monthly hires are missing "beyond the decline observed in the least exposed occupations", i.e. around 55,000 per year (ActuIA calculation). Contract terminations also fall, by about 3,500 per month: the authors speak of an "asymmetric freeze". On job destruction, they write that French data "do not currently allow us to assert that such destruction is under way".

30-39 year-olds almost as affected as 21-29 year-olds

At the occupation level, the decline goes beyond beginners. The gap between the most exposed and least exposed quintiles reaches 20.8 points among 21-29 year-olds and 18.6 points among 30-39 year-olds. It falls to 15.3 points between 40 and 49. After 50, it is only 3.8 points.

IT weighs heavily in this decline, without explaining it alone. Among developers and IT engineers aged 21-29, permanent-contract hiring is about 42% below its November 2022 level, compared with 11% in the four less exposed quintiles. Without them, hiring in the most exposed quintile still falls by 24.5%, compared with 7.5% in the least exposed. According to the note, "IT explains less than a fifth of the gap".

An exposure index scored by Claude Opus 4.7 agents

The index starts from France Travail's ROME 4.0 directory: 1,584 job sheets, 16,740 skills. Claude Opus 4.7 agents score each of them from 0 to 5 on five dimensions, against the capabilities of generative AI as of April 2026. The score starts from the AI's autonomy on the task. It is reduced multiplicatively by tacit expertise and by physical or relational friction.

The 310 occupations are divided into quintiles "without weighting by employment": 62 per quintile, regardless of their weight (ActuIA calculation). The index shows a correlation of 0.56 with the one published by Antonin Bergeaud in 2024. The working paper reports neither human verification of the scores nor comparison with another model. The 17 to 25% figure does not depend on this index: it relies on reported usage.

Economic conditions, interest rates, remote work: other explanations

In its study Les métiers cadres porteurs (2026 edition), Apec (France's executive employment agency) counts 31,330 developer job postings on its site in 2022, 22,400 in 2023 and 12,690 in 2025, i.e. -59.5% in three years (ActuIA calculation). The document does not attribute this decline to generative AI. On 4 June 2026, in an interview with Blog du Modérateur, Véronique Torner, president of Numeum (France's digital industry association), judged the decline in digital recruitment intentions "above all due to the economic situation and lack of business".

The authors examine these avenues. They cite the European Central Bank's interest rate hikes between July 2022 and September 2023, which they say weighed on tech sector hiring. Exposed occupations also lend themselves to remote work: once this criterion is taken into account, the link between exposure and decline among juniors "weakens sharply", according to work by Lambert and Schindler cited in the note. Their answer lies in age. The business cycle "struggles to explain why the gap we observe narrows sharply among older workers", they write.

Denmark, United States, Norway: what other studies measure

In Denmark and the United States, gaps point in the same direction as in France. In Norway, the gap between occupations is not significant.

The Danish central bank also comments on the fate of candidates not hired. In its press release of 16 September, its governor, Signe Krogstrup, concludes that they found employment elsewhere.

At Anthropic, the share of juniors in technical hires fell from 27% to 9%, according to internal data published on 16 October. They do not specify the period or the definition of "junior". On 30 June, a French Treasury note judged that the labour market entry of young people "seems to be slowing".

What the authors recommend to employers

The note makes two recommendations to employers, public and private: "redefine entry-level positions around what AI does poorly" and "preserve on-the-job training pathways". It also proposes an indicator to monitor. "Tracking hiring by age, contract and occupation offers an earlier signal", the authors write.

Another signal concerns apprentices. Among 21-29 year-olds in the most exposed quintile, the share of apprentices rises from 6% at the end of 2019 to 10% at the end of 2023, according to the authors' calculations. The movement began before generative AI: the number of apprentices in France more than doubled between 2017 and the end of 2023, according to Insee (France's national statistics institute) cited in the note. Apprenticeships "may already be playing a shock-absorber role", the authors write. Their effectiveness "will be judged by the hiring rate of apprentices at the end of their contract", which the note does not provide.

The next measurement will come from the Base tous salariés (all-employee database) for 2024. It will make it possible to check "whether the freeze in flows is already beginning to show up in stocks", the authors write.

Add ActuIA to my preferred sources on Google

Our articles will then appear first in Google Top Stories.