AI Jobs Impact Stays Modest, Echoing Computer and Internet Eras

The aggregate numbers have barely moved

Nearly four years after ChatGPT launched, an argument is firming up among economists. Generative AI has spread quickly through U.S. workplaces, but the wider labor market looks much as it did in earlier technology transitions. It has not seen a sharp break.

Morgan Stanley Research made this case in an April analysis. It concluded that AI's effect on employment has so far been modest, with little sign of broad job losses. Where displacement does show up, it is concentrated among younger workers in highly automatable roles.21 The bank's long-run view is optimistic. It expects AI to raise productivity and real wages over time, as earlier innovation waves did after an initial period of disruption. It also warns that faster adoption focused on replacing workers could widen inequality.21

The most systematic tracking points the same way. The Budget Lab at Yale updated its monitoring in October. It found that the occupational mix is not yet changing in ways that clearly line up with AI's arrival. Measures of AI usage show no link to changes in employment or unemployment.11 The researchers' earlier work drew the historical comparison directly. They measured how fast the mix of jobs shifted after ChatGPT and set that against the years after 1984, when commercial computing took off, and after 1996, when the commercial internet arrived. The post-2022 pace ran only about one percentage point above the internet-era path. Much of that shift was already under way in 2021, before generative AI tools were widely available.12 For scale, the Budget Lab pointed out that even during the internet boom, the occupational mix in 2002 differed from 1996 by at most around seven percentage points.12

The Peterson Institute reached the same conclusion in its review of the research. Disruption from AI so far has not outpaced recent technological change, and the reshuffling of occupations has not sped up since ChatGPT's release.23

Adoption is broad, but the payoff stays with individuals

What makes this picture odd is how quickly workplace AI has spread. Gallup reported that half of employed Americans now use AI in their roles at least a few times a year. Thirteen percent use it daily.1 By the second quarter of 2026, 47% of U.S. employees said their employer had integrated AI tools to improve productivity, efficiency or quality, up from 41% the quarter before.6 Among leaders, frequent use rose from 17% in 2023 to 51%.5

The productivity reports show a consistent split. Workers say the tools help them personally, but few see their organizations changing. In AI-adopting organizations, 65% of employees told Gallup that AI had improved their productivity. Only about one in ten strongly agreed that it had transformed how work gets done.1 Glean's Work AI Index found the same gap. Digital workers estimate that AI automation saves them about 11 hours a week, yet only 13% say their organization performs significantly better as a result.4 McKinsey's global survey found that 80% of respondents credit AI with improving their individual productivity. Meanwhile, the share of "AI high performers" attributing meaningful earnings to the technology has stayed flat at about 6%.10

One likely reason the gains stall is the extra work AI creates. SHRM found that workers estimate saving about six hours a week with AI. But 65% also report spending moderate to significant effort reviewing and editing AI output, and only half feel confident in what the tools produce.2 Boston Consulting Group reported that close to half or more of employees spend extra time reviewing, correcting or directing AI output.9

The type of tool also matters. Gallup's data show the strongest reported gains among people using AI for coding assistance or process automation, where 77% report a positive effect. Writing (68%) and search (65%) rate lower. Workers who use AI for seven or more purposes report gains 90% of the time, compared with 45% for those using it for one or two.65 PwC found that the most AI-ready companies are 2.2 times as likely to redesign workflows around AI rather than simply add tools.8 Taken together, this suggests that workflow automation, not chat assistants added on top of old processes, is where the productivity story will be settled.

This pattern fits the historical comparison. Economists have long argued that general-purpose technologies such as electricity and computing paid off only after organizations rebuilt their processes around them. The current data look like the early stage of that cycle: wide individual use, narrow organizational change, and few signs yet in the labor statistics.

Where the evidence splits: young workers

The main disagreement is about entry-level employment. A body of U.S. research finds real damage. A Census-based replication of Stanford's "canaries" work found that in the most AI-exposed fifth of industries, hiring of 22-to-24-year-olds fell about 9% right after ChatGPT's release and has not recovered.29 A companion study of 6.7 million graduates found that those from the most exposed majors, mostly computer science, became 5 percentage points less likely to be employed right after graduating. Their starting earnings fell about 13%, a hit comparable to graduating into a recession.29 Across 41 countries, multinational affiliates that adopted generative AI reduced the junior share of their workforce by 1.9 percentage points.29

Other researchers dispute the timing. One study found that unemployment risk in AI-exposed occupations began rising in early 2022, before ChatGPT existed.29 An Economic Innovation Group paper found that job postings in exposed occupations started falling in 2022 too, which lines up better with rising interest rates than with large language models.23 Population-wide data from Norway and Finland show little or no AI-linked harm to young workers.29 Brookings noted that remote work and pre-existing trends are strong alternative explanations for the overall declines.25

Firm-level data complicate the story further. Ramp economists linked actual AI spending to workforce records at 21,000 U.S. firms. They found that heavy adopters grew employment 10%, and entry-level employment 12%, relative to firms that adopted later.29 A Harvard Business School analysis of almost all U.S. job postings found a split result. Openings in routine, automation-prone roles fell 13% after ChatGPT, while demand for analytical, technical and creative roles grew 20%.24

The most defensible reading is that both sides are partly right. Overall stability coexists with displacement in specific groups. Young workers in exposed fields appear to be bearing real costs, though some of those costs started before generative AI and are tangled up with the 2022 rate shock.

Wages may move before jobs do

One underappreciated finding suggests the labor-market effect may show up in pay rather than headcount. An Apollo white paper used actual usage data from the Anthropic Economic Index rather than theoretical exposure scores. It found that real wage growth in high-exposure occupations ran 6.7 points slower after 2023, with no detectable change in employment.30 The effect was concentrated at the bottom: a 24.3% drop for service workers and 10.7% for the lowest wage quartile.30 Other studies cut against this. One working paper found small positive wage effects with no significant job losses in exposed occupations.22 A Danish study linking adoption to administrative records found null effects on earnings and hours.29 The evidence on wages is therefore unsettled, but researchers who focus only on job counts may miss the effect.

Why history is a guide, not a guarantee

The historical comparison is reassuring, but it has a limit. A Brookings Papers analysis drawing on two centuries of innovation argues that AI differs from past waves because it mainly substitutes for cognitive tasks. On that basis, the authors expect it to shift demand toward occupations with lower education and lower wages, reversing the long pattern of technology favoring educated workers.28 McKinsey's projection is more upbeat. Based on historical rates of occupation creation, it estimates AI could generate roughly 500,000 to two million jobs in occupations that barely exist today.26

Executive expectations are also running ahead of the evidence. Only 14% of McKinsey respondents said AI shrank their workforce in the past year, less than half the 32% who had predicted it would. Even so, 39% now expect headcount declines in the coming year.10 Yale's researchers have stressed that their findings describe the past and do not predict the future.19

On the current record, AI has not yet broken the pattern set by earlier technology transitions. Individual productivity is rising, organizations are lagging behind, and the job mix is shifting slowly. The warning signs are concentrated in early careers and possibly in wages. Whether the old pattern holds probably depends less on how capable the models become than on whether employers use automation tools to redesign work or mainly to cut costs.