Artificial intelligence increases productivity by 4% in European companies that adopt it and pushes their wages upward, without net job destruction in the short term. In the United States, 750 executives surveyed by the Atlanta Federal Reserve confirm this trend: widespread productivity gains, few layoffs, but major internal reorganizations. The first convergent data from both sides of the Atlantic paint a nuanced picture that contradicts both the prophets of mass unemployment and the evangelists of frictionless growth.
The question is no longer whether AI destroys jobs — it reconfigures them by first slowing entry and recruitment for entry-level positions rather than through broad contraction. The question is who guides this transition and how companies finance their employees’ skills development.
The Essentials
- European companies adopting AI see their labor productivity increase by 4% on average according to the European Investment Bank’s survey of 12,000 firms
- In the United States, 64% of 750 executives surveyed by the Atlanta Federal Reserve report productivity gains linked to AI, with acceleration expected in 2025-2026
- Companies that frequently use artificial intelligence are 4% more likely to hire than those that use it more rarely
- Employment of 15-29 year-olds declined by 7.4% in computer activities, 5.8% in publishing, and 3.7% in management consulting activities
- Positions requiring AI skills command a 56% wage premium compared to comparable roles
Adoption Progresses, Gains Confirm
The Federal Reserve Bank of Atlanta’s survey of nearly 750 American business executives provides the first systematic assessment of the economic impact of generative AI. 64% of respondents report measurable productivity gains, compared to 52% in the first survey wave in 2024. This progression reflects both improved tools and organizational learning: companies are better at integrating AI into their processes.
Sectors differ markedly. Financial services and insurance lead the way with 78% of executives reporting significant gains, followed by tech (71%) and business services (69%). Manufacturing and construction show more modest rates around 45%, reflecting more recent adoption and use cases still in development.
On the European side, the European Investment Bank’s survey confirms these trends across a sample of 12,000 companies. Firms that have adopted AI record an average labor productivity increase of 4%, with peaks at 7% in information-intensive services. This transatlantic convergence suggests that observed effects stem not from national specificities but from the technology itself.
Wages Rise, Employment Reconfigures Differently
Contrary to catastrophic predictions, AI adoption comes with higher pay for those who master these tools. Positions requiring AI skills command a 56% wage premium compared to comparable roles according to PwC’s 2025 Global AI Jobs Barometer. The European survey reveals that employees in AI-adopting companies earn on average 3.8% more than their counterparts in non-adopting firms, all else equal. This wage premium reflects several mechanisms: sharing of productivity gains, attraction of qualified talent, and upgrading of positions.
In the United States, 58% of executives report increasing headcount over the past 12 months, compared to just 14% who reduced it. Among the latter, only 23% attribute job losses to AI — others cite classical macroeconomic factors. “AI enabled us to manage more clients with the same teams, but we haven’t fired anyone,” testifies a management consulting firm executive in the Fed survey.
This apparent stability of aggregate employment masks, however, significant internal reorganizations. In France, employment of 15-29 year-olds declined by 7.4% in computer activities, 5.8% in publishing, and 3.7% in management consulting activities, suggesting that employment adjustment relative to AI first materializes through slowed entry and recruitment for entry-level positions. 42% of European companies using AI report significant evolution in their employment structure, even without changes in total headcount. Administrative and data entry positions decline while analysis, supervision, and customer relations functions grow.
The situation in South Korea, where AI has actually created free time for employees without job destruction, illustrates this logic of transformation rather than substitution.
Tasks Shift More Than Jobs
Granular data analysis reveals that AI transforms job content more than it eliminates jobs. According to the Fed survey, 71% of executives observe substantial evolution in their employees’ tasks, even when job titles remain unchanged. Administrative assistants spend less time formatting documents and more time validating automatically generated content. Financial analysts shift from data compilation to interpreting predictive models.
This transition materializes in European figures: an 18% drop in time spent on repetitive tasks in adopting companies, offset by a 22% increase in time allocated to creative and relational activities. “Our employees spend less time searching for information and more time transforming it into recommendations,” explains a banking sector executive interviewed by the EIB.
A particularly revealing Harvard Business School study shows that machine learning-trained AI succeeded in reproducing 71% of trading decisions by traditional fund managers. This is no longer prospective. It’s a concrete signal that high-value cognitive tasks are within AI’s current scope.
The most exposed professions don’t disappear uniformly. Office employees specializing in data processing see their role evolve toward verification and enrichment of automated content. Sales staff use AI to personalize their approaches but retain the relational dimension. Lawyers delegate legal research but keep strategy and advocacy.
This logic of complementarity rather than substitution explains why aggregate employment resists better than expected. AI increases the processing capacity of existing employees before replacing them.
Retraining, the Linchpin
Maintaining employment depends massively on organizations’ ability to train their employees in new tools. 68% of American adopting companies launched AI training programs in 2025, versus 31% the previous year. These programs cost an average of $2,400 per employee according to Atlanta Federal Reserve estimates.
In Europe, 59% of AI-using firms report increasing training budgets by more than 15% in 2025. This increase covers both technical training on AI tools and development of complementary skills — critical thinking, creativity, relationship management — that technology doesn’t replace.
Approaches differ by sector. Banks favor internal training with 3-6 month curricula. Consulting firms outsource through partnerships with specialized platforms. Industrial companies rely on learning by doing with gradual support.
But this commitment could erode. 34% of American executives believe their training budgets cannot keep pace with technological evolution beyond 2026. This tension foreshadows a possible shift: if productivity gains become commonplace while adaptation costs remain high, the incentive to train could give way to replacement logic.
This dynamic echoes tensions observed with telework, which worsened youth unemployment by privileging experience over training.
The Gap Widens Between Adopters and Laggards
AI’s competitive advantage creates growing market polarization. Companies that master these tools attract top talent by offering higher wages and career prospects. Those who delay struggle to retain qualified collaborators.
This dynamic shows through in European data: the productivity gap between adopting and non-adopting companies widened by 2.1 points in 2025, after 1.4 points in 2024. Front-running firms strengthen their position while laggards accumulate losses.
The phenomenon particularly affects SMEs. 43% of European companies with fewer than 50 employees still haven’t adopted generative AI tools, compared to 12% of firms with over 1,000 employees. This fracture stems from learning costs, integration complexity, and lack of internal expertise.
The consequences appear on the labor market. The advantage strongly favors those who train before it becomes an imposed constraint. Employees at large equipped companies develop valuable AI skills, while those in non-adopting structures see their qualifications depreciate. This asymmetry feeds a talent concentration that could heighten sectoral and territorial inequalities.
The global digital divide observed between developed and emerging countries reproduces itself at the company level within advanced economies themselves.
From Assistance to Autonomy, the Real Rupture Approaches
Current data capture AI in its assistance phase — tools help humans be more productive without replacing them. But transformation first strikes codified cognitive tasks, junior positions, support functions. In France, 27% of tasks could be automated by 2030. 47% of American executives anticipate their AI tools will handle complex tasks without human supervision by 2027.
This transition will radically alter observed equilibriums. When AI shifts from assistance to autonomous execution, productivity gains will no longer necessarily translate into wage increases but potentially into staff reductions. The first affected sectors will be those where AI already masters basic tasks: customer service, accounting, data analysis.
A Coface study forthcoming shows that one job in six (5 million jobs in France) will be automatable within three to four years. Certain job families like law, finance, and computing are most exposed. Early signals are emerging. 23% of technology companies surveyed by the Fed are already experimenting with AI systems capable of generating code without human intervention. In finance, autonomous trading algorithms gain sophistication. The legal sector tests AI capable of drafting standardized contracts.
This evolution questions current retraining models. Training employees to use AI becomes less relevant if AI becomes autonomous. The challenge shifts to skills technology cannot replicate — creativity, empathy, ethical judgment, leadership.
The State Faces Its Responsibilities
Convergent data from both sides of the Atlantic outline a major political challenge. AI creates value and maintains employment short-term, but at the cost of rapid reorganization requiring massive support. Companies finance this transition today opportunistically — they find it in their interest. This logic could reverse when technology gains autonomy.
Governments unequally anticipate this shift. The European Union launched an AI training program in 2025 funded with 3.2 billion euros over three years, targeting employees in the most exposed sectors. The United States favors tax incentives to companies investing in retraining, but without binding results.
China experiments with a more dirigiste approach with AI training quotas by sector and sanctions for failing companies. This diversity of approaches will constitute a real-world test of public policy models facing technological disruptions.
The window for action remains open. The coming years are a transition window. What’s at stake now is who emerges strengthened and who suffers. Current data show AI can create shared growth if the transition is guided. But this possibility will last only through the assistance phase. When autonomy takes over, today’s choices will determine whether this technology benefits the many or deepens inequalities.
Sources
Sources: