IMF AI Jobs Warning Puts Entry-Level Work on the Front Line

The "tsunami" warning comes back, with a new focus

Kristalina Georgieva used the opening of the IMF–World Bank Annual Meetings in Bangkok to warn again that artificial intelligence is reaching the labor market like a "slow-moving tsunami." This time she put young people at the front of the line.2122 The IMF managing director said that people entering the workforce usually start in routine, repetitive roles and work their way up. If those roles disappear, she argued, a career path breaks, and she remarked that no young graduate pictures a barista job as their professional future.2124

The figures she gave matched what the press reported. The IMF expects about 40% of jobs worldwide to be "augmented, downgraded or eliminated" over time. Exposure is roughly 60% in advanced economies and 23% in low-income countries. She stressed that the lower figure for poor countries means AI has not reached them yet, not that they are safer.218 One discrepancy runs through the coverage. The IMF's original January 2024 research put low-income exposure at 26%, not 23%, and found that about half of exposed jobs in rich economies could gain from AI rather than lose out.269

The metaphor is not new either. Georgieva first used it in May 2024 and repeated it at Davos in January 2026. At Davos she said AI tends to remove the tasks that make up entry-level jobs first.3028 The Bangkok version changed the context around the warning more than the warning itself.

The "first rung" problem

Singapore President Tharman Shanmugaratnam, who co-chairs the World Bank's Jobs Council, gave the warning a local example. He said demand for graduates' first jobs is already shrinking. He urged governments to make "crisis-era" support measures permanent and to give firms incentives to hire new graduates.2129 Singapore's own data support his point. About 74% of 2025 university graduates in the labor force found full-time work, down from about 88% of the 2022 cohort. The share unemployed and looking for work rose to about 10%, from under 4%.26

The mechanism matters for anyone following workplace AI tools. Junior jobs in research, administration and data processing are the work that current AI products are built to absorb.12 Market research on AI productivity tools lists email triage, document processing, meeting transcription and summarization among the main enterprise uses. It identifies meeting assistants, AI search and workflow automation as growth categories.16 Taking minutes, drafting the follow-up email and pulling notes into an action list used to be routine training work for new hires. Copilot in Teams, Gemini for Workspace and Zoom's AI Companion now promise to do much of it automatically.20 One industry roundup describes a financial services group that used Copilot for note-taking and cut documentation time by nearly 30%.20 Each case like that is a small productivity gain. Added together, they may remove part of the entry-level workload.

In this reading, the danger for young workers is not that a robot takes a whole job. Employers get the same output from fewer junior hires, and the apprenticeship built into routine work disappears. Shanmugaratnam's description of displacement arriving "task by task" fits that picture. He argued that this pace gives workers some time if they can retrain on the job.29

Where the IMF's own evidence is weaker than the headline

Read side by side, the coverage shows a gap between the IMF chief's framing and the Fund's own economists. At a separate Bangkok session, IMF senior economist Divya Kirti presented research showing that demand for AI skills is about twice as high in heavily exposed occupations as in less exposed ones. Hiring demand for programmers fell about 10% relative to less exposed jobs over 2024–2025.4 Kirti said directly that the study has not found clear evidence that younger or entry-level workers are hit harder than experienced staff. It also has not found clear wage effects. In her reading, employers are not dropping humans but favoring candidates who can use AI.4

Outside research is divided too. A Harvard working paper found junior employment fell about 9% at firms that adopted generative AI, compared with firms that did not. A May 2026 study from Warwick, the LSE and Oxford found that much of AI's apparent effect on junior hiring faded once exposure to remote work was taken into account.26 Goldman Sachs research in August found AI-related labor pressure was concentrated among entry-level workers, and that job-opening growth has been weaker in AI-exposed industries since 2022.75

Some outlets reported the Bangkok remarks as AI "wiping out" 40% of jobs, or claimed that automated workflows are "aggressively" removing junior roles.25 That overstates the case. The 40% figure measures exposure, which includes jobs that AI makes more productive, not confirmed losses.13 Still, the uncertainty does not make the warning hollow. A drop in graduate employment, a measurable fall in hiring for exposed occupations and a strong employer preference for AI skills all point the same way, even if no single study isolates AI as the cause. On balance, the IMF is right to raise the alarm early, but its strongest claims rest on a plausible mechanism rather than proven displacement.

Adoption is broad but shallow

The workplace data suggest the disruption is real but slower and messier than vendor marketing suggests. Gallup's May 2026 survey found that 52% of U.S. employees use AI at work in some form, 30% use it at least a few times a week and only 15% daily.15 Germany's Konstanz study found worker adoption rising only from 35% to 38% in a year. Only 55% of users said their most-used tool was officially provided by their employer, which suggests much of the adoption is informal.14 Gallup also found that only about one in ten employees at adopting organizations strongly agree that AI has changed how work gets done.12

The productivity record is just as uneven. Among employees at AI-adopting organizations, 65% said AI improved their productivity. That figure rises to 90% among people who use it for seven or more purposes.11 Self-reports are weak evidence, however. One cross-survey review notes they show perceived benefit, not measured output.12 An NBER survey of 6,000 executives in four countries found nearly 90% reported no measurable productivity or employment gains from AI over three years.30 ActivTrak's tracking found that after workers adopted AI, time spent on email, messaging and business management rose and no activity category fell. In other words, the tools added work rather than replacing it.17 Elsewhere, 45% of U.S. workers said they had to fix or redo a colleague's AI-assisted work.17

Software development is the clear exception. A JetBrains survey of more than 15,000 professional developers found that 90% use AI coding agents at work weekly and 68% use them daily.19 It is probably no coincidence that programming is also the occupation where the IMF measured a relative drop in hiring demand.4 Coding appears to be the field where the link from adoption to fewer hires is easiest to see, and other knowledge work may follow it.

Growth now, a fiscal problem later

Georgieva placed the jobs warning inside a broader economic outlook. She said AI investment is acting as both a demand shock and an investment shock, helping hold global growth near 3% despite an energy supply shock. She estimated AI adoption could add 0.5 to 1 percentage point to growth in advanced Asia and 0.2 to 0.8 points in emerging Asia.2129 The coverage disagrees on how to read this. Market-focused analysis treated it as support for tech-linked Asian assets and for interest rates staying higher for longer.26 Others noted that the headline growth figure leans heavily on data-center spending concentrated in a few economies.30

The longer-term problem is taxation. If AI shifts income from labor to capital, governments that rely on taxing wages could lose their main revenue source. Georgieva put the question bluntly: what do we tax?2223 She also said AI is creating lower-paid work in construction, data centers and services while routine work shrinks, an outcome she called "not a story to celebrate."2324 World Bank President Ajay Banga was more upbeat. He argued that cheap "small AI" tools could create jobs in middle- and lower-middle-income countries, where 1.2 billion young people are approaching working age.2125

What to watch

The practical question for employers is not whether to use meeting assistants, copilots and workflow automation. Most already do. The question is whether they redesign junior roles so new hires still learn the judgment that the routine work used to teach. Bangkok participants raised the same point, warning that graduates need room to develop critical thinking and the ability to check AI output.4 If companies treat AI only as a reason to cut junior headcount, the "first rung" concern will become a measurable trend. Governments, meanwhile, are moving from asking whether AI will disrupt work to deciding how to soften the impact.21 Whether "crisis-era" support becomes standard policy, as Tharman proposed, may be the clearest sign of how seriously they take the threat.