The scene that follows is made up, assembled from the categories an ECB survey uses. Nobody by this name exists. Marta is 24, with a degree in business and a temporary contract at a logistics firm outside Valencia. Three days a week she asks an assistant to draft her supplier emails and clean her spreadsheets, and she reckons it saves her three hours. In June her manager, who uses the same tools more than anyone in the office, tells her the permanent post she was promised is “on hold until the autumn.” Nobody has said that the assistant is the reason. Nobody has said it is not.

Her story sits inside a set of numbers that Philip Lane, a member of the European Central Bank’s Executive Board, read out at a dinner in Dublin on 13 October. Speaking at the International AI Summit, he said that the share of euro-area workers who use AI at work has doubled in two years, from 26 per cent in 2024 to 52 per cent now. In the same breath he said that “we do not yet see a major change in aggregate employment trends but hiring plans are muted, which especially affects the prospects for new graduates.”

Adoption is wide and shallow

Lane drew on two ECB surveys. In the enterprise survey for the fourth quarter of 2025, 64 per cent of firms said they used AI experimentally or moderately, and only 7 per cent called their use significant. The household survey supplied the 52 per cent. Between those two figures lies the gap the speech keeps returning to: almost everyone has touched the tools, and few organisations have rebuilt anything around them.

The bank’s staff published the worker figures in more detail on 26 August. António Dias da Silva, Laura Lebastard and David Sondermann draw on the Consumer Expectations Survey, which polls roughly 20,000 people in 11 euro-area countries each month. Adoption stood at 41 per cent in 2025. It reaches 61 per cent among highly educated workers and 37 per cent among those with less schooling, and younger workers are about 20 percentage points more likely to use AI than older colleagues. The median user reports saving three hours a week, 7.7 per cent of working time. Because only 48.8 per cent of workers report using AI and saving time, the authors put the economy-wide gain nearer 3.8 per cent, and they add that saved hours become output only if workers and employers put them to use.

The savings are lopsided by task. Generating or debugging code yields nearly eight hours a week, but only about 8 per cent of workers use AI for that. Writing, research and text editing are the common uses, and they save far less. Lane’s speech makes the same point, that time savings are “highly concentrated in specific sectors and specialised tasks such as coding.”

Why the aggregate numbers stay flat

A paper presented at the Brookings Papers on Economic Activity this spring offers a caution against reading the survey too quickly. Alexander Bick, Adam Blandin, David Deming, Nicola Fuchs-Schündeln and Jonas Jessen combined worker and firm surveys from 2025 and 2026. They found that industries with higher AI adoption saw faster productivity growth in recent years, a relationship that was statistically significant and similar in size in Europe and the United States, though the authors say they do not establish causality. They found no clear evidence that industry-level adoption is associated with employment changes. They also found that adoption tracks personnel practices, and in particular whether firms actively encourage workers to use AI.

Lane’s own caveat is that it is “too early to detect any material impact on aggregate productivity.” The ECB blog post points to published estimates of extra annual productivity growth from AI over ten years that range from 0.1 per cent to 3.4 per cent, and says further ECB work puts the euro-area figure at around 0.35 percentage points a year.

Where the young workers are

Aggregate employment can hold steady while the entrance narrows. Two earlier pieces of evidence, both covered on this site this week, point at the same door. The Federal Reserve’s October Beige Book reported that five of its twelve districts had firms that reduced staffing through attrition, delayed hiring or expected less job creation after raising productivity with AI. A Stanford Digital Economy Lab paper, revised on 12 August and built on ADP payroll data through June 2026, finds no sign of economy-wide displacement but employment of workers aged 22 to 25 in AI-exposed occupations 19 per cent below where it would have been, working mainly through reduced hiring. And Anthropic’s own economists reported that junior technical hires at the company fell from 27 per cent to 9 per cent of hires, while its agents worked six times as many hours as the humans directing them. Each of these comes from the United States or from one company. Lane’s speech carries the same pattern into euro-area territory, in a single sentence of survey findings.

He lists what would follow if substitution advances: weaker bargaining power for workers, slower wage growth and a lower labour share of income. Workers in sectors that serve the AI buildout are, in the ECB’s surveys, more optimistic than those in knowledge-intensive sectors facing substitution.

Token prices against wages

The line in the speech with the most economic content is short. “The net impact on employment will also depend on the evolution of labour costs versus token prices.” A central banker is saying, in public, that the question of whether a task goes to a person or a model is a price comparison, and that the price of the model side is a number that moves.

Lane places Europe behind in the race. Over the last two years, digital investment in the United States has grown twice as fast as in the euro area, and American workers spend two to three times as much of the working week using AI as workers in the largest euro-area economies. The Stanford AI Index, he says, counts 59 notable AI models from the United States and 35 from China, and one each from France and the United Kingdom. The United States hosts 75 per cent of global AI computing capacity and Europe 5 per cent. He still calls AI already “a primary factor driving current activity levels”, with firms devoting around 10 per cent of their investment to it this year and AI-related lending making up roughly a quarter of recent credit growth to firms. He adds that the global AI boom is plausibly contributing to higher long-term interest rates, which he calls a material tightening of financial conditions for the euro area.

The speech also leaves things unsaid. Survey answers are self-reported, and the speech gives results without methods. The ECB’s own data show no aggregate employment effect yet. Lane makes no forecast of how many graduate jobs are at stake, and the sources do not supply one.

The invented office, a year on

Return to Marta’s office, which exists only here. Suppose that next spring her manager opens two figures side by side: what an hour of a junior’s time costs, and what the equivalent work costs in tokens. According to Lane, that comparison, repeated across thousands of managers and firms, is what will settle the question the surveys cannot yet answer. The first figure is set by contracts and wages, while the second is a price that can move. Whether the post she was promised comes off hold depends on which of the two figures the manager trusts.

References

  • Bick, A., Blandin, A., Deming, D. J., Fuchs-Schündeln, N., & Jessen, J. (2026). Mind the Gap: AI Adoption in Europe and the U.S. NBER Working Paper 34995. nber.org
  • Brynjolfsson, E., Chandar, B., & Chen, R. (2026, revised 12 August). Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence. Stanford Digital Economy Lab. digitaleconomy.stanford.edu
  • Dias da Silva, A., Lebastard, L., & Sondermann, D. (2026). AI adoption and the productivity promise: what workers report. The ECB Blog, European Central Bank. ecb.europa.eu
  • Federal Reserve System (2026). Beige Book, October 2026: National Summary. federalreserve.gov
  • Lane, P. R. (2026). The AI opportunity for Europe. Dinner remarks at the International AI Summit, Dublin, 13 October. European Central Bank. ecb.europa.eu
  • Wilmers, N., Catipon, R., Massenkoff, M., & McCrory, P. (2026). How employment is changing at Anthropic. The Anthropic Institute. anthropic.com