AI and Jobs
Research report · 16 October 2026
Ten leading living economists on AI, the global economy and jobs
The evidence does not show AI causing an economy-wide collapse in employment. It does show a warning sign for young workers in jobs where AI can do the tasks itself. The bigger questions are still open: will AI produce lasting productivity growth, will new work replace the work it displaces, and will wages keep pace with profits? The ten economists below disagree on all three. They agree on one thing: the outcome depends on how firms use AI and how governments manage the transition.
Shortfall in employment of 22–25-year-olds in highly AI-exposed US jobs, against peers in less-exposed jobs. Descriptive, not causal.
Brynjolfsson et al., Aug 2026Share of jobs facing potential automation by generative AI: low- and middle-income vs high-income countries. Not an unemployment forecast.
World Bank, Aug 2026What US firms expect AI to do to their productivity and their own employment over the next three years. So far, almost no realised effect.
Bloom, San Francisco Fed, Mar 2026The ten views
Grouped into three camps. Select a name to read the full profile and its 2026 sources.
1 Philippe Aghion 2025 Nobel laureate; economist of growth and innovation Cautious optimist: jobs are lost and created, and policy decides the net Growth optimists
Aghion is cautiously optimistic about growth and total employment. In the CEPR paper he and his co-authors argue that counting automatable tasks as lost jobs is a mistake. It ignores the tasks that remain in each occupation and the extra demand higher productivity creates. They see no strong conceptual reason to expect AI to reduce total employment. They do expect job losses to arrive before replacement jobs, and they note weaker early-career hiring in exposed roles in France, Denmark and the US.
Jobs and policy: Danish-style “flexicurity”: income insurance combined with well-targeted retraining. Add to that education that builds adaptability, and a European industrial policy that steers AI towards uses that complement workers. In Barcelona he argued that AI also speeds up the generation of ideas, and so growth. A reported estimate of 0.68 percentage points a year of extra business-productivity growth is a model projection, not a measured gain, and the CEPR argument does not depend on it.
2026 sources: Aghion, Bergeaud, Bunel and Roulet, Labour market policies in the age of AI, CEPR, 28 September 2026; Barcelona address, June 2026.
2 Joel Mokyr 2025 Nobel laureate; economic historian of innovation Optimist: AI speeds science and helps with labour shortages Growth optimists
Mokyr is broadly optimistic about AI’s power to speed scientific discovery and, over time, to improve health, energy and living standards. He also says plainly that no one knows AI’s ultimate effect. His 2026 remarks are a historian’s informed judgement, not a quantitative forecast.
Jobs: He expects labour shortages to be a major challenge over the next 20 years, and thinks AI could take on tasks fewer people are willing or able to do. This bears most on ageing societies and occupations short of workers. It does not mean workers in exposed office jobs will escape disruption. His 2026 view comes from a news-agency report of a public appearance, which is weaker than a signed paper.
2026 sources: HINA report of Mokyr’s remarks at the EUonAIR AI Summit, 6 July 2026.
3 Michael Spence 2001 Nobel laureate; economist of growth and the developing world The gains depend on AI spreading beyond tech and finance Growth optimists
Spence’s central question is diffusion. AI’s gains depend on whether it spreads beyond technology and finance into construction, health care, retail, hospitality and education. If it does not spread widely, he says, there will be no productivity surge, and the gains will tend to increase inequality. He thinks the US technology sector may be overinvesting in AI, and that governments, not only private firms, should set the agenda.
Jobs: He resists simple forecasts. AI sometimes removes jobs that need high expertise and sometimes jobs that need little. He says the debate often ignores basic economics, such as how demand responds when prices fall and incomes rise. Some traditional jobs will be automated, while other industries see rising incomes and demand (Xinhua).
Global angle: Frontier AI research is concentrated in the US and China, but the tools can be used anywhere. He praised China’s systematic push to apply AI across industry under its 15th Five-Year Plan (Xinhua). In an October 2026 interview he again cast AI mainly as a productivity tool rather than a wholesale replacement for workers, provided diffusion is managed well (ET Conversations, 4 Oct 2026, episode summary).
At the Tsinghua forum, he expected AI to lift labour productivity substantially but said it was too early to know how much that would ease global debt pressures. The same account says he saw emerging economies as generally more optimistic about AI than advanced economies, where job replacement is a more prominent concern.
2026 sources: Tsinghua PBCSF Global Finance Forum’s account of his 18 May remarks, published 22 May 2026; Xinhua interview at the same forum, 20 May 2026; The Edge Singapore, remarks around the Asian Monetary Policy Forum, 29 May 2026.
4 David Autor MIT labour economist whose research shaped the study of automation, trade and wages The question is the value of expertise, not the number of jobs Pro-worker if steered
Autor contrasts the coming AI shock with the China trade shock he helped document. Like that shock, it is big and can happen fast. Unlike it, AI’s effects will be spread across the country rather than concentrated in manufacturing towns. It will reshape tasks rather than wipe out whole industries. And firms will welcome it as a productivity gain rather than suffer it as competition. That last difference, he notes, does not make it good for workers.
Jobs: For Autor the key question is the value of human expertise, not the number of jobs. AI could let mid-level workers such as nurses, designers and production managers do work that once needed specialists, raising their pay. “Inexpert” service jobs, such as care work, security and food service, may stay poorly paid because a large pool of workers can do them. The joint paper takes this further: only AI that creates genuinely new, valuable tasks is unambiguously “pro-worker”, because a tool can raise output while lowering what a worker’s existing skills are worth.
Policy: He favours “universal basic capital”, such as government-seeded baby bonds, as a hedge for people whose only asset is their skills. He sees a universal basic income as politically unviable. The joint paper adds public investment in AI for health and education, tax and competition reform, and a stronger worker voice in technology design.
2026 sources: Noema interview, 16 January 2026; Acemoglu, Autor and Johnson, Building pro-worker artificial intelligence, Hamilton Project, February 2026.
5 Daron Acemoglu 2024 Nobel laureate; economist of institutions and technological change Sceptic of the hype; wants AI steered towards good jobs Pro-worker if steered
Acemoglu remains the leading sceptic of both AI hype and the jobs-apocalypse story. He calls the pitch of AI agents as one-to-many replacements for workers “a losing proposition”. Jobs bundle many tasks, about 30 for an X-ray technician, and AI is poor at moving between them. AI also still lacks easy-to-use apps of the kind that spread word processors and spreadsheets, one reason it has had no seismic effect on jobs or productivity so far.
Jobs and global economy: In his World Bank keynote he put “good jobs” at the centre of the growth question: more output does not guarantee higher wages or a career ladder. Automation has held down wages for many US groups since about 1980 and reduces labour’s share of income, and AI framed around replacing workers could speed that up. Developing countries often lack digital infrastructure and have little say in AI’s direction. His hopeful scenario combines pro-worker AI at the frontier with adequate infrastructure, and he points to India’s investment as a possible model. He also worries that AI firms are hiring in-house economists to shape the story in their favour.
2026 sources: MIT Technology Review interview, 11 May 2026; World Bank ABCDE keynote, 16 June 2026; the joint Hamilton Project paper.
6 Simon Johnson 2024 Nobel laureate and former IMF chief economist AI risks repeating industrial-era mistakes unless steered Pro-worker if steered
Johnson’s historical worry is that productivity gains can come first and shared prosperity never follow. He compares today’s AI incentives with early mechanisation, which displaced workers before new roles developed and fed inequality and unrest. The analogy illustrates a risk; it is not a forecast that history will repeat on the same scale.
Jobs and policy: He singles out entry-level and knowledge work as exposed. With Acemoglu and Autor he argues for AI that raises the value of workers’ expertise. To get it, he proposes building AI expertise in government, public grants for pro-worker tools, stronger competition and changes to tax incentives. At the paper’s launch he said AI is not yet on that path.
2026 sources: Technovation interview, 1 January 2026; the joint Hamilton Project paper; Hamilton Project launch event recap, 27 February 2026.
7 Joseph Stiglitz 2001 Nobel laureate; economist of information, inequality and public policy Bubble and displacement first, “intelligence assisting” later Disruption first
Stiglitz sees two linked risks. AI investment may be a bubble, because it assumes both that AI succeeds technically and that competition stays limited. If the boom reverses, a conventional downturn would follow. At the same time AI may displace routine white-collar work. His statement that AI investment drove roughly a third of last year’s US growth is an interview estimate, not a settled national-accounts figure.
Jobs: He doubts the US has the retraining, active labour-market policy and industrial strategy to move displaced people into good work, much as it lacked a way to move surplus farm workers during the Depression. In the longer run he favours “intelligence assisting”: AI that helps teachers, plumbers and others do better work while human judgement remains central. In his account a bubble and a useful technology can coexist.
2026 sources: Fortune interview, 8 March 2026.
8 Esther Duflo 2019 Nobel laureate; development economist AI may remove the service-job ladder in emerging economies Disruption first
Duflo focuses on the development ladder. In Brazil, service jobs are how many people move from low to middle incomes. She warns that AI could remove that ladder if it replaces office and other service work. Banerjee, interviewed alongside her, gave accounting and legal research as examples and expects office work to decline faster than manual work.
Global angle: Duflo also sees an opportunity. AI translation could lower language barriers and let Brazil compete for international service work now dominated by English-speaking countries. Her view is two-sided: AI may widen inequality inside developing countries, but access to useful tools could also create new cross-border work.
2026 sources: Folhapress interview with Duflo and Abhijit Banerjee in São Paulo, 26 March 2026.
9 Kenneth Rogoff Harvard economist and former IMF chief economist Job losses likely to outweigh gains; countries will drift apart Disruption first
Rogoff’s main concern is that countries will drift apart. Those with a place in the AI supply chain, such as the US, China, South Korea, Japan and Taiwan, will collect the profits and taxes needed to cushion job losses. Countries outside it may face displacement without the means to respond. He sees India’s outsourcing industry as especially exposed, though India has the talent to become a major winner. Mineral exporters such as Chile and the Democratic Republic of the Congo could gain, but only if institutions spread the revenue to ordinary households.
Jobs: His view is harsher than the optimists’. In the ACCA interview he said he expected “job losses are going to be much bigger than the growth gains”, and a backlash larger than the one against globalisation. That is his judgement under high uncertainty, not a measured effect. In April he added a fiscal point: AI is likely to raise capital’s share of income and cut labour’s, so tax revenue grows more slowly than output, and AI will not rescue rich-country budgets.
2026 sources: “The AI economy’s permanent underclass”, Project Syndicate, June 2026 (also published by The Guardian, 2 June 2026); “Will AI Solve Rich Countries’ Debt Woes?”, Project Syndicate, April 2026 (signed republication, 26 April); ACCA outlook interview, January 2026.
10 Paul Krugman 2008 Nobel laureate; trade economist and macroeconomist Turned pessimistic: AI investment crowds out jobs Disruption first
Krugman’s view shifted during 2026. In June he was sceptical that AI would hurt workers much. By October he called his view “much more pessimistic”, and his worry had moved from automation to investment. Huge spending on data centres may come at the expense of investment in more labour-intensive activities. A roughly $8.2 billion AI training campus employs 50 to 150 people once running; the same capital elsewhere would support more than 10,000 jobs.
Jobs: His illustrative calculation suggests that projected data-centre spending through 2032 might support roughly 10 million fewer jobs than other uses of the same capital. He says explicitly that this is not a forecast of 10 million more unemployed people. It is a counterfactual estimate of labour demand, pointing to lower wages and a further shift of income to capital. The result depends on the assumed spending, how much other investment is crowded out, and whether AI’s productivity gains create jobs elsewhere. This argument is in the free part of his essay. Separately, in June he argued that AI’s damage to thinking and learning may matter more than its productivity gains.
2026 sources: “Will AI Crowd Out Jobs?”, 11 October 2026; “What Will AI Do To Our Minds?”, 28 June 2026.
Where the ten differ
Productivity
Aghion and Mokyr emphasise innovation and faster science. Spence says gains depend on spreading AI beyond tech and finance. Acemoglu doubts automation-heavy deployment will deliver gains to match the hype, and Bloom’s survey finds optimism about the future but little realised effect so far.
Employment
Aghion expects job creation to offset losses over time; Mokyr sees labour shortages as the bigger constraint. Rogoff expects much harsher losses. The payroll evidence shows a selective shortfall in early-career jobs, not an economy-wide effect.
Wages and power
Autor, Acemoglu and Johnson argue that more output need not raise the value of labour. Stiglitz stresses institutions to manage the transition. Krugman adds that capital-intensive AI investment may weaken demand for workers even before widespread automation.
Across borders
Spence sees diffusion as the route to shared gains. Duflo sees both threatened service ladders and new export opportunities. Rogoff warns that countries outside the AI supply chain may be unable to pay for protection.
What the evidence shows so far
The ten are describing different mechanisms over different time horizons. Aghion, Mokyr, Spence and Rogoff make long-run or broad predictions. Autor, Acemoglu and Johnson analyse how the design of AI changes wages and the value of expertise. Stiglitz and Krugman add macroeconomic risks from finance and investment. These are not the same kind of evidence and should not be weighed as if they were.
The best direct measurement comes from Brynjolfsson, Chandar and Chen’s payroll study, revised 12 August 2026. Using payroll records for millions of US workers through June 2026, they find no evidence of widespread, economy-wide job displacement. But employment of 22- to 25-year-olds in highly AI-exposed occupations was 19% below where it would have been had it kept pace with less-exposed peers. The gap mostly reflects slower hiring, and it is concentrated where AI substitutes for workers rather than complementing them. The authors call this an early, descriptive indicator, not a causal estimate. Some of the divergence predates generative AI, and it weakens when education is controlled for.
So “little aggregate displacement” and “serious risk to young office workers” do not contradict each other. A few occupations and age groups can suffer while total employment holds steady. Nor does the fact that a task can be automated translate directly into lost jobs: firms can reorganise work, expand output or create new tasks. Aghion’s team stresses those offsets; the payroll data show why exposed groups still need watching.
A firm-level check comes from Nicholas Bloom’s survey, presented at the San Francisco Fed on 24 March 2026. Executives reported almost no realised effect of AI on productivity or employment over the previous three years. US firms expect productivity to rise 2.25% and their own employment to fall 1.2% over the next three. These are expectations, not outcomes, and new firms or jobs elsewhere could offset the losses.
The global distribution of gains
The developing-country picture is more complicated than a single job-loss forecast. The World Bank’s 2026 World Development Report estimates that 4.5% of jobs in low- and middle-income countries face potential automation by generative AI, against 14.2% in high-income countries. It estimates that 16.2% and 18.7% respectively could see meaningful productivity gains. These measure exposure, not predicted unemployment.
Lower exposure overall does not remove the risks Duflo and Rogoff describe. A country may depend heavily on the particular service jobs AI affects, and it may lack the power, connectivity, skills, finance or tax base to capture the gains. Spence’s answer is that deployment, not frontier research, decides who benefits: the tools are open to the whole world, and China’s applied push shows what a deliberate diffusion strategy looks like. Acemoglu adds that infrastructure and a voice in AI’s direction matter as much as the technology itself.
Overall conclusion
As of October 2026, the defensible short-run conclusion is selective disruption, especially in early-career hiring in highly exposed occupations, with no demonstrated economy-wide fall in employment. The medium-run question is whether productivity gains turn into better wages and new careers, or flow mainly to owners of capital. Globally, access to infrastructure and useful applications will decide whether AI narrows or widens the gaps between countries.
The net effect on employment remains unresolved. The common lesson from these ten views is to measure actual outcomes carefully, protect people through the transition, and reward AI that expands what workers can do. GDP growth alone will not settle whether there are enough good jobs.
How the ten were chosen
There is no official ranking of the world’s top living economists, so any top ten is a judgement call. This list weights three things: scholarly distinction (a Nobel prize or equivalent standing), influence on economic policy, and engagement with AI, growth or labour markets. Each person also needs a substantive source published or delivered in 2026. Eight are Nobel laureates in economics. David Autor is the leading economist of technology and work, and Kenneth Rogoff brings an international macroeconomic view. The order is thematic, not a ranking.
Two first choices failed the 2026 rule and were replaced. Our searches found no qualifying 2026 statement on AI by Larry Summers. The 2026 appearances we found for Claudia Goldin covered gender pay and fertility, not AI (Goldin’s media page). They were replaced by Simon Johnson and David Autor. Erik Brynjolfsson was considered. His August 2026 payroll study is the best direct measurement of AI’s effect on jobs, so it appears below as evidence rather than as a profile.
The report gives most weight to the economists’ own papers and essays, then full interviews, then reported remarks. Forecasts, scenarios and observed results are labelled differently. The 2026 rule decides who is eligible; it does not mean every cited prediction has come true. Most of the ten work at US institutions. Spence, Duflo and Rogoff supply most of the perspective from Asia and the developing world.
All sources
Every economist has at least one 2026 source.
Evidence and context:
| Source | Date |
|---|---|
| Brynjolfsson, Chandar and Chen, Canaries in the Coal Mine? (Stanford Digital Economy Lab) | Revised 12 Aug 2026 |
| Nicholas Bloom, The impact of AI on productivity (San Francisco Fed) | 24 Mar 2026 |
| AI offers lifeline to developing economies in an era of weak growth (World Bank) | 4 Aug 2026 |
| All prizes in economic sciences (Nobel Prize) | — |
| Claudia Goldin: interviews, videos and in the news (Harvard) | 2026 |
