Still waters, rapid currents: Early labor market transformation under generative AI | Brookings

Introduction

Think of a marketing professional using a chatbot to draft ad copy, an IT support specialist troubleshooting a user’s problem with one, or a teacher trying to spot homework written by AI. A few years after ChatGPT’s launch, generative AI has quickly become part of everyday white-collar work. Has it changed how much workers earn, or how many people firms hire? Our research provides one of the first comprehensive answers, based on surveys of 25,000 Danish workers across 11 highly exposed occupations, linked to their monthly administrative records through December 2024. The headline result is a precise null: no effects on earnings or hours, either for workers who use chatbots or for the firms that adopt them. Beneath this still surface, however, work is already being substantially reorganized.

Experts disagree sharply about how generative AI will affect work, both in direction and in size. Anthropic CEO Dario Amodei (2025) warns that much white-collar work could be displaced within a few years, while Nobel laureate Daron Acemoglu (2025) expects modest productivity gains spread over decades. The stakes are especially high for young workers: is this the worst or the best time to start a career in an AI-exposed occupation? In an influential paper, Brynjolfsson, Chandar, and Chen (2025) document that early-career employment has fallen sharply in AI-exposed occupations, which has raised concerns that AI may already be displacing entry-level jobs. By contrast, a more optimistic view, advanced by Autor (2024), is that generative AI may put expertise that used to be scarce within reach of more workers, opening paths to better-paid careers. Until now, there has been little direct evidence to weigh these views against each other.

In this article, we describe how workplaces are absorbing AI chatbots: the employer policies, tools, and training now surrounding the technology, and the new job tasks it is creating. We then ask whether these changes have surfaced in earnings, hours, and employment, with occupational mobility emerging as the notable exception. We conclude by discussing what our findings mean for policy: how statistical agencies should measure AI’s labor market effects, why policymakers should be careful attributing aggregate trends (such as the decline in entry-level hiring) to AI, and why the occupational mobility channel deserves attention.

A clean early window on AI adoption

Denmark provides an unusually clean window into this process. Few workforces have taken up generative AI faster than Denmark’s, where adoption rates are on par with the United States (Humlum and Vestergaard 2025). Its labor market is also flexible: hiring and firing costs are low, much as in the U.S., so firms and workers face few barriers to adjusting employment as technology changes.

Critically, Denmark is unusually well equipped to measure who is actually using a new technology: each resident has an official digital mailbox through which Statistics Denmark can send survey invitations. Working with Statistics Denmark, we fielded two rounds of representative surveys, in late 2023 and late 2024, each inviting 115,000 workers to report on their use of AI chatbots and on changes in how their work is organized. The surveys target 11 occupations that are highly exposed to AI chatbots: software developers, IT support specialists, customer support specialists, office clerks, accountants, financial advisors, HR professionals, legal professionals, marketing professionals, journalists, and teachers. We chose these occupations because they involve tasks where AI chatbots can save workers time, based on the exposure assessments of Eloundou, Manning, Mishkin, and Rock (2024). Together, they make up the parts of the labor market where any effects of AI chatbots are likely to show up first and be largest. The 2024 round, the primary basis for our findings, received about 25,000 valid and complete responses across 7,000 workplaces and measured workers’ use of AI chatbots, employer initiatives such as usage policies, enterprise tools, and training, and changes in the organization of work.

The samples are representative of the underlying workforce on characteristics such as age, gender, experience, earnings, and wealth, and survey responses check out against variables also recorded in the administrative registers. Because respondents are drawn from the registers, every response can be linked to monthly administrative data on earnings, hours worked, and occupations—for the respondents themselves, and for the roughly 3 million other workers at Danish workplaces. The link gives us both sides of the picture: the outcomes that registers record, namely earnings and hours, and the adjustments they miss, from shifting job tasks and work organization to the complementary investments that firms make around new technologies (Brynjolfsson, Rock, and Syverson 2021).

Our analysis then asks one simple question: have chatbot adopters fared differently in the labor market than comparable non-adopters? We compare workers who use AI chatbots to otherwise similar workers in the same occupation who do not, and workplaces with chatbot initiatives to similar workplaces without them, tracking outcomes month by month around the launch of ChatGPT in November 2022. Because the records extend back years before ChatGPT existed, we can check that adopters and non-adopters followed similar labor market trajectories beforehand, so that differences emerging afterward can more plausibly be attributed to the technology.

The currents: How workplaces are absorbing AI

Employers have rapidly embraced AI chatbots. Our 2024 survey asked workers about their employer’s policy on AI chatbot use, with five possible answers: chatbot use is encouraged, use is allowed, use is not allowed, the workplace has no policy, or the worker does not know of one. As Figure 1 shows, 43% of workers in our 11 occupations work for employers that explicitly encourage chatbot use, while only 6% are prohibited from using them. Among workers in encouraging workplaces, 61% have access to enterprise chatbots and 39% have received training in using them. The rapid adoption of AI chatbots has been bottom-up: even in workplaces that take no active steps, about 41% of workers have used chatbots at work. Yet, despite this high worker-driven baseline, adoption more than doubles in workplaces that combine encouragement with enterprise tools and training: 93% of workers in such settings have used AI chatbots at work, 28% use them daily, and 19% report saving more than one hour per day. This pattern, in which adoption and reported benefits are strongest where employers pair the tools with organizational support, echoes a long line of research showing that new technologies pay off most when firms invest in complementary organizational practices, including how they manage people (e.g., Brynjolfsson and Hitt 2000; Bloom, Sadun, and Van Reenen 2012).

This widespread embrace of AI marks a shift from the early days of ChatGPT, when many employers restricted its use over concerns about data confidentiality and output accuracy. Outright bans are now rare, persisting mainly in occupations that handle sensitive data, such as financial advisors, or that require high factual accuracy, such as legal professionals.

AI chatbots do more than speed up existing work; they also create new tasks, shown in Figure 2. Workers who reported new tasks described them in their own words, and we classified these free-text descriptions into six categories: AI ideation, AI content drafting, AI data insights, AI quality review, AI ethics and compliance, and AI integration. Roughly four in ten new tasks involve generating content with AI: the ideation, drafting, and data insight categories. About a third involve overseeing AI, combining quality review of its outputs with ethics and compliance work, such as checking legal accuracy or detecting AI-generated homework. And the largest single category, about a quarter of new tasks, is integration: fine-tuning AI assistants, writing usage policies, and building chatbots into everyday workflows. Even in workplaces with no AI policy, about 8% of users have taken on entirely new tasks, and this share doubles where employers actively support adoption. Most users (85%) spend the time they save on other job tasks; far fewer do more of the same tasks or take more leisure. This is consistent with work on “job transformation” arguing that technologies reshape what jobs consist of long before they eliminate them (Freund and Mann 2026; Autor and Thompson 2025).

The examples, drawn from workers’ own descriptions, are concrete. Marketing professionals report “prompting and iterating with AI to produce marketing copy, social media posts, and product descriptions.” Software developers report “fine-tuning AI coding assistants by providing feedback and project-specific examples.” Legal professionals describe “developing organizational AI usage policies and guidelines.” And a large share of teachers report new work related to “detecting AI-generated homework.”

Notably, the new work extends even to workers who have never used AI chatbots. About 4% of non-users report new workloads resulting from the tools, and the share is highest in workplaces with chatbot initiatives. Among teachers who have never used the tools, 10% report new AI-related work, mostly keeping tabs on, and responding to, students’ AI use. These spillovers to non-users signal a broader workplace transformation, consistent with task-based theories in which new technologies reinstate labor demand by creating new work (Acemoglu and Restrepo 2018).

The emergence of new work is also consistent with the historical record. Research by David Autor and coauthors shows that 60% of U.S. employment today is in job titles that did not exist in 1940 (Autor, Chin, Salomons, and Seegmiller 2024)—the creation of new work is how labor markets have absorbed new technologies over time. What is unusual is catching the process in the act: our data capture the emergence of AI oversight and integration tasks within two years of the technology’s arrival, before any of them appear in official occupational classifications.

The surface: Why wages and headcounts have not yet moved

Have the rapid currents of adoption and reorganization surfaced in what workers are paid? Figure 3 says no. In levels, Panel (a) shows that adopters do earn more than comparable non-adopters (about 9% more where employers encourage use and 4% where they do not) and follow rising earnings trends. But these gaps were already there before ChatGPT. Once we index to ChatGPT’s launch, the difference-in-differences in Panel (b) is flat, and its confidence intervals rule out any differential earnings change larger than 2%, for both workers and workplaces. The contrast between the two panels echoes the debate over computers and wages in the 1990s (Krueger 1993; DiNardo and Pischke 1997): users of a new technology often do earn a sizable premium, yet such premia can say more about who picks up the tools than what the tools do for them.

These null results hold even for daily chatbot users, for workers who save more than an hour a day, for those who took on new AI tasks, and for those whose employers actively support adoption. They also hold for early adopters, already using chatbots when we surveyed in 2023, who have had longer for any effect to appear. And they hold in each of the 11 occupations, software development and marketing included, where decentralized wage setting should let pay track individual productivity most readily.

Firm-level outcomes are just as still. Workplaces that encourage chatbot use, including those that provide enterprise chatbots and training, show no differential changes in headcounts or wage bills, job creation or destruction, or the composition of hires or separations. Incumbent workers at adopting workplaces have not been pushed out of their jobs or their occupations. The workplace estimates are as precise as the worker-level ones, ruling out effects larger than 2%.

Workers themselves agree on this still surface: asked directly, 98% of adopters, and virtually all non-adopters, report that AI chatbots have not affected their earnings, suggesting that spillovers to workers who do not use the tools have also been limited.

Are these null wage effects surprising? In standard economic reasoning, workers should be compensated for productivity gains they bring to the job themselves (Becker 1964), so it is notable that even high-intensity users with large reported benefits see no earnings gains. Rigid wages are a natural explanation for small short-run impacts, but our null results persist even in occupations with flexible pay. In fact, pay in the exposed occupations is far from frozen: among otherwise similar workers, earnings changes over this period vary substantially for reasons unrelated to AI. Pay moves; it just does not move with chatbot adoption. One possibility is that using chatbots functions as a job amenity: workers take the payoff in easier or more engaging work rather than in pay (Rosen 1986). Another is that workers overstate how much the tools actually help them, a key reason we link self-reports to third-party administrative records rather than relying on either source alone.

How do these null effects square with the benefits that users report? The reported time savings, averaged across all users and work hours, amount to a few percent of total work time, and the gains that do materialize appear to be absorbed into the reorganization of work rather than paid out: the workers reporting the largest time savings are also the ones most likely to have taken on new tasks. Read this way, the null results do not signal that nothing is happening. They signal how adjustment is happening: inside workplaces, through the reorganization of work, rather than in the external labor market of wages and hiring. Earlier general-purpose technologies followed a similar script, with the productivity J-curve literature showing organizational investment running well ahead of anything measurable in economic outcomes (Solow 1987; David 1990; Brynjolfsson, Rock, and Syverson 2021).

The exception: Occupational mobility

One recorded outcome does move with chatbot adoption: occupational mobility. To measure it, we take each worker’s latest occupation, the one they held in December 2024, and track how many hours they worked in that occupation over time. Workers who recently switched into it will show a rise in these hours after the switch, while those who never left will not. Figure 4 shows that, compared with similar non-adopters, adopters now put in about 4% of a full-time equivalent more hours in their latest occupation. In other words, adopters are more likely to have switched occupations since the arrival of AI chatbots. Unlike the earnings premia in Panel (a) of Figure 3, this association shows no pre-trend before ChatGPT’s launch, suggesting it reflects the technology rather than adopters simply being more mobile workers.

Where do switchers go? Mostly into IT support and clerical roles, occupations where workers have more freedom to choose their tools and that require no formal credentials. By contrast, we find no excess transitions into licensed occupations, such as teaching and accounting, which require several years of prior education regardless of what tools a newcomer commands. The switchers do well: their earnings growth runs 12 percentage points ahead of that of other Danish workers, as they move into better-paying roles where chatbots are more directly relevant. The link between adoption and switching appears to be driven primarily by individual workers rather than their employers: it strengthens with the intensity of individual use, tripling among daily users and among those reporting large time savings, but is fairly similar across employer chatbot initiatives.

These patterns align with experimental evidence that AI chatbots provide the greatest productivity benefits to workers with less prior expertise (Noy and Zhang 2023; Brynjolfsson, Li, and Raymond 2025), and they support the vision in Autor (2024) of generative AI as a tool that helps workers access otherwise scarce expertise and move into better-paying careers. Occupational switching is also a margin worth watching in its own right: moving to better jobs is a core source of wage growth over workers’ careers (Kleven, Kreiner, Larsen, and Søgaard 2025), so these transitions may serve as early indicators of AI’s longer-run labor market effects. The switchers are still too few to move average earnings among adopters, but if the mobility channel scales as adoption deepens, it is a plausible route by which generative AI’s benefits reach paychecks.

Our mobility results also offer a different perspective on how generative AI is shaping the prospects for launching a career in an exposed occupation. Denmark mirrors the declines in early-career employment documented in the United States: several of our exposed occupations, including software development, legal professions, and marketing, have seen falls in early-career jobs since 2022. But our data let us divide these trends between adopting and non-adopting firms. The share of early-career workers has evolved no differently at workplaces that encourage chatbot use, and the estimates are tight enough to rule out differences larger than a third of a percentage point. In other words, while AI-exposed occupations have seen falls in early-career jobs, firms adopting generative AI are not what is driving them. What does drive the aggregate declines remains an open question, with recent studies pointing to forces beyond firms’ adoption of generative AI, including pre-existing trends in exposed occupations and the rise of remote work (Iscenko and Curto Millet 2026; Frank et al. 2026; Lambert and Schindler 2026). The two sides of our mobility evidence are also worth reading together: using AI chatbots has helped workers step into new occupations, while employers adopting the tools have not yet pushed incumbent workers out of theirs.

What this means for policy

Measuring AI’s labor market effects requires both administrative data and surveys. Administrative data alone, such as the earnings and employment records in the U.S. Census Bureau’s Longitudinal Employer-Household Dynamics (LEHD) program, miss the workplace changes AI chatbots are already producing in job tasks and work organization, and they cannot tell which workers and firms have adopted the tools. Surveys alone, such as the AI questions in the Census Bureau’s Annual Business Survey (ABS) and Business Trends and Outlook Survey (BTOS), may overstate AI’s reach, because the time savings and new responsibilities workers report do not always translate into changed wages or hours. This is why linking the two matters: only by connecting the currents of workplace adjustment to the hard surface of earnings and hours can one get an objective sense of whether, and when, these adjustments translate into market outcomes. Efforts in this direction are welcome and would be well worth expanding. Statistical agencies should track these new margins jointly. Two are particularly important. The first is complementary workplace investments: adoption and its reported benefits peak where employers combine encouragement with enterprise tools and training—the kinds of intangible investments that the productivity J-curve literature identifies as preceding measurable economic gains (Brynjolfsson, Rock, and Syverson 2021). The second is the reorganization of work itself, including the new tasks that technologies create (Autor, Chin, Salomons, and Seegmiller 2024)—adjustments that conventional labor market statistics do not record.

Policymakers should be careful about attributing aggregate trends to AI. It is key to measure the separate impact of adoption, especially in the early years of a new technology, when many forces move labor markets at once. The early-career case illustrates the stakes: aggregate declines in exposed occupations are consistent with AI displacing entry-level jobs, yet splitting the same trends by actual adoption shows that firms using the technology are not driving them. Getting the diagnosis right matters: a policy aimed at AI will do little good if AI is not what is driving the decline. It matters for public sentiment, too: only 18% of young Americans say they feel hopeful about AI, and commencement speakers, former Google CEO Eric Schmidt among them, were booed at this spring’s graduation ceremonies for praising the technology. To the extent these fears rest on the belief that AI is already eliminating entry-level jobs, that belief runs ahead of what our evidence shows firms actually adopting the tools are doing.

The occupational mobility channel deserves attention. Occupational switching is the one margin where AI adoption already registers in administrative records, and moving to better jobs is a core source of wage growth over workers’ careers. Indeed, research by Petrova, Schubert, Taska, and Yildirim (2026) suggests that industry-level exposure to a previous automation technology, industrial robots, has reduced workers’ chances of moving into better-paid occupations, with lasting consequences for their careers. Our evidence concerns a different margin: workers’ own adoption of the tools rather than their industry’s exposure to them. On that margin, generative AI points in the opposite direction, aligning with the vision in Autor (2024) of a tool that can rebuild middle-class career ladders: adopters who switch move into occupations that pay more and where chatbots are more relevant. The pattern of these moves also shows where the channel can operate at all: mobility gains appear where workers can choose their own tools and where formal credentials do not bar entry. Whether this channel grows as adoption spreads is still open for research and policy, which makes occupational transitions worth tracking early.

Conclusion

So far, the labor market record on generative AI resembles still waters running over rapid currents. Workplaces are adopting AI, work is being reorganized, and adopters are moving into new occupations, but earnings and hours have stayed flat. The pattern recalls Solow’s famous 1987 remark about the IT revolution: “You can see the computer age everywhere but in the productivity statistics.” Indeed, the two patterns may be mutually reinforcing: when earnings and hours hold steady, workers and workplaces can absorb technological change by reallocating tasks and making other adjustments that administrative labor market data do not record. If so, the flat earnings and hours are not a sign that nothing is changing. They are a sign that the change is taking place on margins that standard economic statistics miss. Our data open a window into the trough of the productivity J-curve, where work is being reorganized before anything registers in hours or earnings. Seeing into that trough takes infrastructure of the kind we build here: surveys of what is changing inside firms, linked to administrative records of what eventually surfaces in the labor market.

Is this the worst or the best time to launch a career in an AI-exposed occupation? Our results on occupational mobility paint a cautiously optimistic picture: chatbot adopters have shifted into higher-paying occupations where using chatbots is more relevant. Although this is still early evidence, it suggests a somewhat more hopeful picture than the “white-collar bloodbath” sometimes portrayed in public media. We are still in the early innings of generative AI. An important avenue for future research is to track how and when the surface catches up with the currents.