Research opportunity · Aggregate null or distributional shift? Employment
Do company agreements on AI change how firms adopt it and whom they hire?
Earlier version (v1)
This is RO-LD02-06 as it was published at v1, kept unchanged. The current version is v2; its changes are listed there.
Question
In French firms, does signing a company-level collective agreement that addresses AI change the subsequent pattern of hiring (junior versus experienced, AI-related versus other roles) and training compared with similar firms in the same sector that have not signed one, and what do the agreements commit firms to?
Institutional practice unknownOne PhD chapter or paperLow to medium costPartly addressed
Why it matters
The debate's early evidence comes from countries with very different labour institutions, and Norwegian and Finnish authors suggest unions and employment protection may explain why their results differ from the US, without testing it. Adoption effects in the Danish data run mainly through recruitment. If negotiated agreements steer firms toward retraining and away from hiring freezes, the "which margin moved" answer depends partly on labour institutions, not only on the technology. Researchers have catalogued AI clauses, but no study found asks whether they change what firms then do.
The debate lists this among what we still do not know: How do collective bargaining, worker voice and labour regulation alter adoption strategies?
Evidence that motivates it
- Danish firms that began using AI in 2023 had roughly 11% lower relative employment growth by late 2025 than comparable non-adopters, mainly through lower recruitment, while economy-wide employment remained relatively unaffected.
Danish adopters show lower relative employment growth mainly through recruitment, not layoffs; recruitment is therefore the margin on which agreements would need to make a difference, and the claim does not license reading it as job loss. - Exposure, adoption, actual use, task productivity, firm productivity, labour demand, hiring, employment stock and wages are distinct empirical objects; evidence at one stage does not establish the next.
Adoption, use and labour demand are distinct stages; an agreement can act on any of them, so the design measures hiring and training separately from the fact of signing.
A possible design
Two parts. (1) Text-as-data: download French company agreements from the public ACCO open-data base (agreements concluded since 1 September 2017), identify those addressing AI, and hand-code a sample for commitments (consultation before deployment, training, job-security guarantees, limits on monitoring, hiring commitments); validate any automated classifier against the hand-coded sample. (2) Event study: match signing firms to non-signers in the same sector and size band, and compare job postings (junior versus experienced, roles naming AI tools) and, where accessible, administrative hiring and training records, before and after signature. Report the results by type of commitment. Interviews with a few union and HR negotiators can explain the mechanisms.
Data needed
ACCO company agreements (free, published by the French government under the Licence Ouverte; the dataset page does not say whether the files carry firm identifiers or only names, so this needs checking before linkage); job-posting data by firm (commercial postings data, or the national employment service's data if a research agreement can be reached); French administrative firm and worker data through the secure-access centre (CASD) if the project can obtain access.
Feasibility
One PhD chapter or paper. Low for part (1), which uses only open data; medium for part (2), depending on postings-data licences and secure-access fees. Part (1) alone is a strong master's thesis. Signing an AI agreement is not random (larger, unionised firms sign more often), so part (2) compares signers with closely matched non-signers and should describe its findings as associations unless a credible source of variation, such as a sector-level agreement that binds some firms, can be found.
Has this been done?
Partly done already: related work covers some of this question (first-pass screen of 12 October 2026; not yet checked in depth).
Inventories and reviews of AI clauses exist for the US and Europe, including counts of French company agreements mentioning AI, but none of those read links clauses to adoption or hiring outcomes. An Oxford project led by a doctoral researcher (2025–2028) studies how German works councils shape AI adoption with administrative records and new employer surveys, and lists no results yet; it concerns works-council presence rather than agreement content. A Dutch study links firm technology adoption and union strength to re-employment after job loss, for technology in general and before generative AI, and a German firm-panel study finds the fall in continuing training after AI adoption is driven mainly by firms without a collective agreement, but it uses only coverage, not agreement content, and pre-dates generative AI. No study was found that links the content of AI agreements to subsequent hiring and training.
How strong this evidence is
- Who searched and checked: an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Every check so far was done by AI; no person or domain expert has checked it yet.
- What was read: 1 of the 5 nearest works below was judged from an abstract, summary or search listing only; it is marked. Not every entry says how much of the work was read.
- Where the search could not reach: Screened by web search on the date shown, mostly in English with one French query; French-language academic work, Google Scholar, SSRN and paywalled journals were not systematically searched, and one article was read at abstract level only. Absence of a hit is not proof that the work does not exist.
- Before building a project on it, run your own search in a scholarly database such as Scopus, Web of Science, Google Scholar or SSRN.
Nearest work found
- Barop (principal investigator), "Worker Voice and AI Adoption," Oxford Internet Institute research project funded by Schmidt Sciences (November 2025 – September 2028)
Project page: examines whether German works councils steer AI toward complementing workers, combining administrative firm records with new employer surveys. No publications or results listed. Closest in aim; differs in country, institution (works councils, not agreement content) and outcome. - Kerckhofs, "Collective bargaining on artificial intelligence at work," Eurofound (26 September 2025)
Article reviewing 2024 developments in about a dozen European countries; cites a CNAM analysis finding AI mentioned in 242 French company agreements and reports clauses on training, transparency, oversight, limits on substitution and job security. Describes clauses; does not measure effects on adoption or employment. - Kresge, "Negotiating Tech: An Inventory of U.S. Union Contract Provisions for the Digital Age," UC Berkeley Labor Center (last updated 2 July 2025)
Provisions from more than 175 US union agreements, drawn from over 500 contracts, grouped by type. Informational inventory; no link to adoption or employment outcomes. - ten Berge & Lippényi, "Firm-level technology implementation and finding a new job: The moderating role of industry unionization," Economic and Industrial Democracy 47(1), 271–303 (2026), DOI 10.1177/0143831X251326346 Judged from an abstract, summary or search listing only
Abstract only read: Dutch matched employer–employee data 2001–2016; job searches after job loss are longer for workers leaving firms implementing new technology, and the association is weaker in more unionised industries. Technology in general, before generative AI; re-employment rather than hiring or adoption strategy. - Muehlemann, "AI Adoption and Workplace Training," IZA Discussion Paper No. 17367 (October 2024)
German BIBB establishment panel 2019–2021, AI and machine learning in general: AI adoption is associated with less continuing training, mainly in smaller firms and firms without a collective agreement. Uses only whether a firm is covered by an agreement, not what it says, and is associational.
Search queries used
collective bargaining unions works councils generative AI adoption employment effects firms evidence 2026collective agreements clauses generative AI analysis text corpus agreements AI provisions hiringworks council codetermination AI adoption Germany establishment panel IAB employment junior hiringunionization AI adoption firm level empirical paper "union" artificial intelligence adoption employment effects heterogeneity collective bargaining coverageLégifrance base ACCO accords d'entreprise open data intelligence artificielle accords analyseeffect of collective agreements on artificial intelligence on firm hiring training outcomes empirical study company agreements AI clauses France
Provenance
RO-LD02-06 v1. Drafted by Claude (SocioLens Project Lead), AI-drafted from the LD-02 debate on 2026-10-12. Checked by an independent AI reviewer with fresh context (not a person), 12 October 2026: every nearest-work citation checked against its source page; corrections applied. Not yet reviewed by a domain expert. This is a research hypothesis, not a finding.