The Question

A robotic arm working alongside a human on a factory floor, both illuminated by blue light

Every major wave of technology has triggered the same fear: that machines would take all the jobs. The spinning jenny threatened hand spinners. The power loom threatened weavers. The computer threatened typists, bank tellers, and travel agents. In every case, the technology eliminated some jobs, created others, and left the economy larger than before. The optimists have been right — so far.

But there is a serious argument that this wave is different. Previous technologies replaced physical labour or narrow clerical tasks. AI can write, code, analyse, diagnose, advise, and reason. When the machine can do cognitive work, the historical playbook may not apply. By 2033, we predict the net job count from automation will be negative — even as the quality and pay of the jobs that survive and emerge will be higher than what was lost. That is a genuine economic gain and a genuine social crisis, simultaneously.

What the Evidence Shows

The World Economic Forum's Future of Jobs Report 2025, based on a survey of more than 1,000 large employers, expects 170 million jobs to be created and 92 million displaced worldwide by 2030 — a net gain of 78 million. But those totals mask an asymmetry. Among the fastest-declining roles are cashiers and administrative assistants — now joined by graphic designers as generative AI spreads. The growth comes at two very different ends of the labour market: in specialist technology, data and AI roles, and in frontline work such as farmworkers, delivery drivers, care workers and educators. Forty-one percent of employers said they plan to reduce their workforce as AI automates certain tasks. The gain and the loss are often happening to different people.

In the US, economists led by MIT's David Autor have documented decades of "job polarisation": growth in high-skill, high-wage jobs and in low-wage service jobs, while middle-skill clerical and production work shrank. The question is whether AI accelerates that pattern. The first hard evidence suggests it is starting where careers begin. Stanford researchers, using payroll data covering millions of US workers through June 2026, found no sign of economy-wide job losses from AI — but employment of 22-to-25-year-olds in the most AI-exposed occupations, such as software development and customer service, has fallen well behind that of their peers, mainly because companies are hiring fewer young people.

"Employment of young workers (ages 22–25) in AI-exposed occupations now stands 19% below where it would be had it kept pace with that of their less-exposed peers; experienced workers show no comparable gap."

— Brynjolfsson, Chandar & Chen, Stanford Digital Economy Lab, "Canaries in the Coal Mine?", revised August 2026

Not every study sees disruption yet: Yale's Budget Lab found in 2025 that measures of AI exposure showed no sign of being related to changes in overall employment or unemployment. But there is also a speed problem. Even when new jobs emerge from technological change, workers historically needed a generation to retrain and reposition. If AI disrupts labour markets at the pace current trajectories suggest, the transition window is far shorter than anything previous waves allowed. The people losing jobs now may not have a decade to retrain. They may have two or three years.

The new jobs will be better — but better for whom, exactly, is the question that will define the next decade.

Why This Is Happening

AI has crossed the cognitive threshold. For most of computing history, machines could only do what they were explicitly programmed to do. Modern AI — large language models, vision systems, reasoning engines — can generalise. They can handle novel situations, interpret ambiguous instructions, and produce work that requires genuine judgment. It is the difference between a calculator and a colleague.

The economics of automation have inverted. Historically, automating a task required expensive custom engineering. A robot that could sort packages cost millions to design and deploy. Now, a general-purpose AI tool can handle a wide range of knowledge tasks for a modest monthly subscription per user. Earlier waves of automation already left a measurable mark: a landmark study by Daron Acemoglu and Pascual Restrepo found that each additional industrial robot per thousand workers reduced the US employment-to-population ratio by about 0.2 percentage points and wages by about 0.42%. The economic threshold for replacing a human worker has dropped dramatically — and will continue to drop as the technology improves and prices fall.

Education and retraining systems are not built for speed. Community college retraining programs run one to two years. Corporate training budgets are typically the first thing cut in a downturn — precisely when retraining is most needed. The infrastructure for helping workers adapt is slow and underfunded.


What Could Happen

The Polarisation Deepens Most likely

By 2033, the job market has bifurcated sharply. High-skill workers who can collaborate with AI — managing, directing, and correcting it — command premium wages and face strong demand. Routine cognitive and manual workers face sustained displacement without an obvious landing spot. A large population of working-age adults cycles through gig work, retraining programs, and underemployment. GDP grows. Inequality grows faster. The political consequences are significant.

Managed Transition With Policy Support Possible

Governments intervene at scale: universal retraining credits, shorter working weeks to spread available work more broadly, stronger wage floors, and investment in human-intensive sectors — care work, education, community services — that AI cannot easily replicate. The displacement is real, but it is absorbed into an expanded social infrastructure rather than dumped onto individuals. This requires a level of political coordination that is rare but not unprecedented — the post-war welfare state was built in a decade.

AI Creates a New Job Explosion Less likely

AI productivity gains are so large that they generate entirely new industries at a scale that absorbs the displaced workforce rapidly. Just as the internet created social media managers, SEO specialists, and app developers from scratch, AI spawns categories of work we cannot yet name. Wages broadly rise, including for lower-skill workers, as the economic pie expands faster than any cohort can be excluded from it. This is the optimistic scenario, and it requires unusually fast new-industry formation.

Our Assessment
We assign 67% probability — more likely than not that by 2033, automation will eliminate more routine jobs than it creates, while the jobs it creates pay significantly more. This is not a verdict on whether AI is good or bad. It is a verdict on timing and distribution. The new jobs will exist — but the gap between when old jobs disappear and when new ones become accessible to the workers who lost them will cause real suffering at scale. The critical variable is not technology. It is whether society chooses to build the bridges between the jobs that are disappearing and the jobs that are emerging — or leaves workers to swim across on their own.

What Can We Do

Person at a desk using an AI interface on a screen, working alongside digital tools

Whether automation is a threat or an opportunity for you depends enormously on where you sit in the labour market — and what you do now.

Identify the automatable parts of your job. Be honest. Routine, rule-based, repetitive tasks — regardless of how skilled they feel — are the most vulnerable. The parts of your job that require genuine human judgment, relationship management, ethical reasoning, or creative synthesis are far more durable. Know which is which.

Learn to work with AI, not just alongside it. The workers who will fare best are not those who ignore AI or those who are replaced by it — they are those who become skilled at directing, correcting, and amplifying AI tools. This is a learnable skill. Start now, before the pressure is acute.

Invest in uniquely human capabilities. Empathy, persuasion, physical dexterity in complex environments, creative vision, and ethical judgment are all things AI does poorly. Skills that concentrate in these areas — nursing, teaching, skilled trades, leadership, design — are more durable than skills in pure information processing.

Do not wait for your employer to retrain you. Corporate retraining programs are underfunded and often too slow. Online learning platforms, community colleges, and professional communities offer faster, cheaper, and more current skill development.

Support policies that spread the gains. The productivity gains from AI will be enormous. How those gains are distributed — through wages, public services, or shareholder returns — is a political question. Engaging with that question is not abstract. It will directly determine whether the AI revolution is experienced as liberation or displacement by the majority of workers.