French companies have begun to measure what artificial intelligence brings them. What they don’t show is who they pass it along to. And without institutional architecture to enforce it, they never will be.

In the most exposed sectors, productivity has nearly quadrupled

In financial services or software publishing, productivity nearly quadrupled between 2018 and 2024, rising from 7% to 27% [1]. Six years. Sixty percent of executives surveyed in France said their company had achieved significant operational productivity gains thanks to AI [2].

These figures are real. What happens next is equally real: who captures the value thus created, and how. Productivity is measured in company accounts. Its distribution, by contrast, is traced nowhere in any systematic way.

In 2024, France recorded only a very weak increase in labor productivity, barely +0.4% [3]. The United States posted +2.2% the same year [3]. The gap doesn’t stem from an absence of gains in the most advanced French companies, but from their concentration and the slowness of their diffusion. In 2023, labor productivity in the eurozone had fallen by 0.9%, its steepest decline since the 2008 financial crisis [4]. When the aggregate stagnates, it masks a growing dispersion between leading-edge companies and the rest of the economy.

The precedent of the steam engine, and its seventy-year lag

The question of distribution is not new. In reality, it is the oldest question in industrial economics. The first industrial revolution multiplied the wealth produced. But it was only in the twentieth century, with the Fordist compromise and the welfare state, that workers began to receive their share [5]. According to Axelle Arquié, an economist at CEPII, this delay cost nineteenth-century Britons about seventy years of impoverished lives before benefits began to spread in earnest [5].

Arquié sees in AI a comparable shock [6]. This time, automation targets managers and the highly educated. Without institutional compromise, productivity gains will once again be absorbed by capital.

The mechanism she identifies is precise. AI fragments intellectual labor just as mechanization had transformed the craftsman into a factory worker. The machine executes tasks, the human verifies results. Sociologist Juan Sebastián Carbonell calls this “augmented Taylorism” [6]: a gradual deskilling, without salaries bearing any trace of it.

One agreement in a thousand mentions AI

The numerical finding is unambiguous. Between September 2017 and April 2024, among the 285,421 collective company agreements published on Légifrance, 242 make reference to AI [7]. One in a thousand.

A few large companies (Orange, AXA France, BPCE, Decathlon) have opened negotiations. These texts remain highly marginal. No dynamic of technological social dialogue has emerged at the level of the French economy [8].

This absence is not for lack of tools. French law on profit-sharing has existed for decades. In 2024, 54% of private-sector employees were covered by at least one employee savings scheme [9].

But this coverage concentrates on large structures. In companies with more than 1,000 employees, the rate reaches 89.1% of workers [9]. In very small and small businesses with fewer than 50 employees, it falls to 21.1% [9]. Yet it is precisely there, in SMEs and mid-size firms, that AI is beginning to transform daily tasks—with no redistribution mechanism in place.

The problem lies in the base. Existing schemes measure and redistribute accounting profits, not productivity gains tied to a particular technology. In a standard profit-sharing agreement, no one knows what share of results is due to deploying an AI tool. No one has any obligation to calculate it or negotiate it.

Why the market doesn’t correct itself alone

AI can amplify human capabilities. But market failures push it elsewhere: misaligned incentives among firms and developers, path dependence, a systemic preference for automation over capability enhancement [10].

The logic is straightforward. Automating a task immediately shows up in costs. Training an employee to work with AI takes time, generates uncertainty, and produces benefits the firm doesn’t capture alone. Companies perceive greater returns in deploying technologies that automate expertise rather than in creating new tasks or valorizing employee skills [10].

“We are currently not on the trajectory of pro-worker AI,” Simon Johnson sums up [10]. This finding, formulated from the United States, applies identically to France.

French budgetary constraint adds a tension. Tax breaks on profit-sharing (taxes the state forgoes in the name of value-sharing) already represent several billions. Their extension to new AI-related schemes will require an explicit trade-off in coming budgets. Conditioning what the state already foregoes costs less than creating a new expense, but requires defining what is expected in return.

Linking tax benefits to negotiation on AI gains

The lever is not to create an AI bonus. That would be a one-off payment without changing who decides what. The trade-off is to make mandatory, in any company deploying AI systems in its activities, negotiation over the sharing of productivity gains attributable to those tools. This obligation would change who sits at the table, and on what basis each calculates what everyone should receive.

Concretely: any profit-sharing or employee participation agreement signed after deploying an AI tool would include a productivity indicator linked to that use: verifiable, enforceable, included in the calculation base. Employee representatives would have audit rights over the gains measurements thus produced. Companies that don’t negotiate would lose access to tax and social exemptions applicable to profit-sharing.

This is what the industrial revolution failed to do for seventy years and what the Fordist compromise ultimately imposed: institutionally linking the sharing of gains to their measurement [5]. Among the nine levers Acemoglu, Autor, and Johnson identify for steering AI toward workers, the reexamination of taxation to rebalance capital and labor is the one directly touching the architecture of collective agreements [10]. After this reform, companies deploying AI could no longer benefit from tax advantages of value-sharing without first having negotiated over the share attributable to those tools.

Sources

[1] PwC, “AI Jobs Barometer: Rise in Productivity in Sectors Most Exposed to AI,” May 2024, https://www.pwc.fr/fr/espace-presse/communiques-de-presse/2024/mai/ai-jobs-barometer.html (accessed 10/10/2026).

[2] IBM, “The Race for ROI: Study on AI Productivity Gains in France,” November 2025, https://fr.newsroom.ibm.com/Etude-IBM (accessed 10/10/2026).

[3] OECD, “OECD Compendium of Productivity Indicators 2026,” July 2026, https://www.oecd.org/en/publications/oecd-compendium-of-productivity-indicators-2026_734a5e68-en (accessed 10/10/2026).

[4] OECD, “OECD Compendium of Productivity Indicators 2025,” June 2025, https://www.oecd.org/content/dam/oecd/en/publications/reports/2025/06/oecd-compendium-of-productivity-indicators-2025_f1a7de9f/b024d9e1-en.pdf (accessed 10/10/2026).

[5] Axelle Arquié and Thomas Grjebine, “From Industrial Revolution to AI, These Transformations That Shake Up Work (and Its Remuneration),” World Economy 2025, CEPII / La Découverte, December 2024, https://www.cepii.fr/CEPII/fr/publications/em/abstract.asp?NoDoc=14216 (accessed 10/10/2026).

[6] Axelle Arquié, NVO interview, “The Effects of AI Will Be Felt on Employment When the Technology Matures,” April 29, 2026, https://nvo.fr/axelle-arquie-les-effets-de-lia-se-feront-sentir-sur-lemploi-quand-la-technologie-sera-mature/ (accessed 10/10/2026).

[7] IRES, Working Paper No. 03-2025, “Social Dialogue, Collective Agreements and Artificial Intelligence,” June 2025, https://ires.fr/publications/documents-de-travail-de-lires/n03-2025-dialogue-social-accords-collectifs-et-intelligence-artificielle/ (accessed 10/10/2026).

[8] CEET / Employment Knowledge, cited by CE Expertises, “AI in Collective Agreements,” October 2024, https://www.ce-expertises.fr/actualites/lintelligence-artificielle-dans-les-accords-collectifs-etat-des-lieux-et-perspectives/ (accessed 10/10/2026).

[9] DARES, “Employee Savings in 2024,” data cited by Empruntis, August 2026, https://www.empruntis.com/rachat-credits/actualites/2026/epargne-salariale-record-2024-21048/ (accessed 10/10/2026).

[10] Daron Acemoglu, David Autor, Simon Johnson, “Building Pro-Worker Artificial Intelligence,” The Hamilton Project / Brookings, NBER Working Paper No. 34854, February 2026, https://www.hamiltonproject.org/publication/paper/building-pro-worker-ai/ (accessed 10/10/2026).