Reading time: 9 min
The Fund’s own research paints a more complicated picture than the headline suggests — and the workers caught in the middle know it better than anyone.
Kristalina Georgieva does not mince words. At the World Economic Forum in Davos this January, the IMF Managing Director described artificial intelligence as a force bearing down on employment worldwide. At the AI Impact Summit in New Delhi a month later, she repeated the warning. The Fund’s research estimates that roughly 40% of jobs globally — and 60% in advanced economies — will be affected by the technology in the coming years, whether through enhancement, elimination, or outright transformation.
For Europe, the numbers are even more striking. The UK government’s own assessment, published in January 2026 and drawing on IMF methodology, found that approximately 70% of British workers hold roles containing tasks AI could potentially perform or enhance. That figure exceeds the US equivalent of around 60%, largely because the UK economy is more heavily weighted towards the service-sector occupations where AI capabilities overlap most with human work.
We should be clear about what “affected” means here. The IMF’s framework distinguishes between roles where AI is likely to complement workers — boosting productivity and potentially raising wages — and those where it is more likely to substitute for human labour, depressing demand and pay. Roughly half of the exposed jobs fall into each category. The problem is that the headline figure travels without that nuance, and public anxiety fills the gap.
The People Training Their Own Replacements
Georgieva’s macroeconomic framing acquired a sharply human dimension this week. A series of first-person accounts from workers directly involved in training AI systems revealed a pattern of disillusionment that economic models cannot capture.
One editor, who corrects academic papers for non-native English speakers, described being recruited to train what she was told were “assistant editors.” She spent months feeding corrections into the system before the company revealed it was an AI programme. Her fee was subsequently cut. She now earns less correcting the machine’s output — a task she says takes longer than editing from scratch — while catching errors the model produces, from unnecessary punctuation to nonsensical changes in country names.
A marketing writer with industry awards spent months building AI workflows and documentation, believing he would oversee the system. He was made redundant. His former workload is now handled by junior staff using the documentation he created. A translator reports spending four years training AI engines that his employer intends to deploy as a cost-cutting replacement for human linguists.
These are not abstract displacement risks. They are people who were asked to participate in their own professional obsolescence — often without being told that was the purpose of the exercise.
“Training your robot replacement feels like digging your own digital grave.”
— Award-winning marketing writer, laid off after building AI workflows (The Guardian, Feb 2026)
Not everyone is pessimistic. A palliative care consultant who helped pilot an AI chatbot for patients with metastatic cancer noted that it managed roughly half of responses correctly, but still required heavy adaptation and could not replicate the non-verbal cues — body language, facial expressions, tone — that define good clinical care. A mathematics professor working with AI theorem-proving acknowledged that the technology is advancing rapidly, but felt insulated by his teaching role and public-sector employment. Both cases suggest a pattern: the more a role depends on interpersonal nuance, physical context, or institutional relationships, the harder it is for AI to displace.
The Numbers Behind the Noise
Workplace anxiety around AI is measurable and rising. ManpowerGroup’s 2026 Global Talent Barometer, based on interviews with nearly 14,000 workers across 19 countries, found that regular AI usage jumped 13% in 2025, but worker confidence in the technology fell 18%. Baby boomers reported the sharpest decline, at 35%, while Gen X saw confidence drop 25%. Almost two-thirds of the workers surveyed said they were choosing to stay in their current roles despite burnout and dissatisfaction — a phenomenon the report characterises as “job hugging,” driven by fear of what automation might mean for their next move.
The layoff data tells a more complicated story than either side of the debate usually admits. Consulting firm Challenger, Gray & Christmas recorded approximately 55,000 AI-attributed job cuts in the United States during 2025, out of a total 1.17 million layoffs — the highest annual figure since the pandemic year of 2020. AI-linked losses have climbed sharply: through the first seven months of 2025, around 10,000 cuts were tied to the technology, but by year’s end the total had surged past 54,000 — a fivefold acceleration in the second half of the year.
Yet several credible voices urge caution. Yale University’s Budget Lab analysed US labour market data from November 2022 to mid-2025 and found no substantial acceleration in the rate at which the occupational mix was changing — in other words, the composition of the workforce had not yet shifted dramatically since ChatGPT’s launch. Deutsche Bank analysts warned in January that “AI redundancy washing” — companies citing AI as a convenient justification for cuts driven by broader economic pressures — would become a significant feature of 2026. Sander van ‘t Noordende, CEO of Randstad, the world’s largest staffing firm, told CNBC at Davos that the link between those 55,000 cuts and AI was being overstated.
Forrester offered perhaps the most sceptical assessment: it estimates that just 6% of US jobs will be genuinely automated by 2030, and projects that many AI-attributed layoffs will ultimately be reversed as companies discover the technology is not ready to fill the roles it was supposed to replace.
Young Workers Are Feeling It First
If there is one area where the data is starting to converge, it is entry-level employment. Research from the Dallas Federal Reserve, drawing on Stanford University analysis, found that workers aged 22 to 25 in the most AI-exposed occupations experienced a 13% employment decline since late 2022. The drop was driven not by layoffs, but by a reduction in the rate at which young people entering the labour market found work in those roles at all. For occupations with lower AI exposure, entry rates held steady.
This aligns with what the IMF itself has flagged. At Davos, Georgieva noted that entry-level tasks are disproportionately vulnerable to automation because they tend to involve the kinds of structured, repetitive cognitive work that current AI handles well. Meanwhile, workers in roles that have been enhanced by AI — around one in ten positions in advanced economies, according to the Fund’s estimates — tend to earn more and spend more, creating downstream demand for lower-skilled service jobs. The uncomfortable implication is a labour market that hollows out in the middle: AI-augmented professionals at the top earning more, service workers at the bottom sustained by that spending, and a shrinking pool of mid-level and entry roles caught in between.
“The middle class, inevitably, is going to be affected.”
— Kristalina Georgieva, WEF Davos, January 2026
What Actually Needs to Happen
The IMF’s own policy prescriptions centre on skills investment, social safety nets, and regulatory frameworks that can keep pace with the technology. The Fund published an AI Preparedness Index covering 125 countries, which measures readiness across digital infrastructure, human capital, innovation capacity, and governance. Wealthier economies tend to score better, but with wide variation — and scoring well on preparedness is not the same as acting on it.
The practical challenge is timing. AI’s capabilities are advancing faster than most institutions can adapt. Georgieva was blunt about this at Davos: the technology is moving too fast for policymakers to keep up, and the gap between deployment and regulation is widening. The EU’s AI Act provides a regulatory framework, but it was designed around risk categories that may not map neatly onto the messy realities of workplace disruption — an editor unknowingly training her replacement does not fit cleanly into “high risk” or “low risk.”
Denmark’s flexicurity model — which combines relatively easy hiring and firing with generous unemployment benefits and active retraining programmes — is frequently cited by the IMF as a template. Whether it can scale to the speed and breadth of AI-driven change remains untested. What does seem clear from the IMF’s own research is that economies with stronger social protections, higher educational mobility, and more flexible labour markets will navigate the transition more smoothly than those without.
The Honest Assessment
We are likely in a period where the fear of AI displacement is running ahead of the measurable reality — but the measurable reality is catching up. The Yale data and Forrester projections suggest the labour market has not yet been fundamentally reshaped. The worker testimonies and Dallas Fed research suggest the early effects are real, concentrated among the young and the mid-career professionals closest to the training process.
Georgieva’s tsunami metaphor may be premature as a description of what has already happened. As a warning about what the next few years could bring if investment in skills, regulation, and social protection does not keep pace with deployment, it is hard to dismiss. The workers who have spent the past year training AI systems and watching their own roles diminish would probably say the water is already rising.
Sources: Fortune, Business Today, GOV.UK, IMF, ManpowerGroup, CNBC, Dallas Fed, The Guardian, Tom’s Hardware, IMF SDN
