Don't blame AI alone for the two-speed jobs market

THE LAUNCH of ChatGPT in late 2022 was supposed to usher in a jobs apocalypse. The bosses of the firms building large language models have every incentive to sell that story: their valuations depend on the assumption that AI will automate a substantial share of human work. A few months into 2026, the evidence tells a different one. The real story is not mass replacement. It is a quiet divergence between the white-collar workers whose jobs are being reshaped by AI and the rest of the labour market, which is struggling for reasons that have little to do with algorithms.

Indeed, a job-search platform, publishes regular analyses through its Hiring Lab. Its latest data show that AI-related hiring in the UK is growing while broader recruitment remains subdued. About 5.6% of UK job postings mentioned AI or related tools, according to Indeed data from April 2026, a share that has been climbing steadily. The UK is an outlier among peer economies in the intensity of AI's presence in vacancies, yet overall job postings remain 19% below their pre-pandemic level, according to Indeed's 2026 UK Jobs and Hiring Trends Report. The platform's researchers warn that, left unaddressed, a mismatch between the jobs that are growing and the skills and location of available workers could push unemployment to 8% by 2040. That figure is a projection, not a forecast, but it captures a genuine risk: the divergence may widen.

The two-speed metaphor has some purchase. One speed is the knowledge-work economy, where AI is augmenting some tasks and automating others. The other is the wider labour market, where hiring has been soft for well over a year. But the two trends should not be conflated. The UK's job market was weakening at the tail end of 2025. Tax changes announced in late 2024, including a significant rise in employers' social-security contributions and a large increase in the minimum wage, have kept firms cautious. Uncertainty around the government's workers' rights bill and tariff volatility have reinforced the mood. Unemployment edged to 5% in the third quarter of 2025, the highest level since early 2021. Wage growth remains robust but is easing. These are cyclical and policy-driven headwinds, not technological ones.

The deeper question is what AI is doing to the jobs that are not disappearing. A report from the British Progress think-tank in April 2026 analysed Annual Population Survey data covering 412 UK occupations and found no evidence that AI has replaced jobs at scale. Employment trends in the most AI-exposed occupations have not diverged from those in the least exposed. But there is a wrinkle. Employer payroll data from the Office for National Statistics show that wages in high-exposure occupations have grown more slowly since 2019 than in low-exposure ones. The trend predates ChatGPT, which means it cannot be easily blamed on generative AI. It does, however, point to a longer-running pressure on routine white-collar work that AI is likely to amplify. Within the most exposed occupations, the picture is mixed. The number of roles for programmers and finance analysts has continued to grow since 2022, while administrative and clerical positions have contracted. The same degree of AI exposure can produce different outcomes depending on whether a job's structure lends itself to augmentation or replacement.

The incentive structure behind this divergence is straightforward. Firms that adopt AI are looking to reduce costs and increase throughput, not necessarily to fire people. Goldman Sachs, a bank, estimates that generative AI could displace 6-7% of the US workforce if widely adopted, but that the impact on overall employment would be modest and temporary as new jobs emerge. Stanford's Institute for Economic Policy Research (SIEPR) reached a similar conclusion in May 2026: AI's effect on overall employment is likely small, though it may contribute to a tougher market for new graduates. The bank's economists argue that historical technological transitions have typically raised the unemployment rate by 0.3 percentage points for every 1 percentage point of technology-driven productivity growth, with the impact usually disappearing after two years. The lesson is not that disruption will be painless. It is that it will be selective and transitory, at least by historical standards.

The trouble is that transitory disruption still requires institutional adaptation. The UK's service-sector economy means that around 70% of workers are in occupations containing tasks that AI could potentially perform or enhance, according to IMF estimates cited in the UK government's own January 2026 assessment. That is a higher share than in the US or other advanced economies, reflecting the UK's service-heavy composition. About half of these exposed workers are in roles where AI is more likely to boost efficiency; the other half are in roles where it may displace. The government has established an AI and Future of Work Unit to address the gap in evidence. That is a step in the right direction, but it is a small one.

The real bottleneck is not AI capability-it is adaptation. Indeed's projection of rising unemployment by 2040 rests on three converging trends: the retirement of baby boomers, slower immigration, and AI reshaping white-collar work. The first two are demographic and policy choices, not technological outcomes. The UK has tightened immigration rules, reducing the flow of foreign workers. At the same time, the retirement of an entire cohort is removing experienced workers faster than new entrants can replace them. AI is then reshaping the types of skills that remaining employers need, creating a mismatch that the education system is ill-equipped to address quickly.

To be sure, the case for alarm has been overstated. The hype cycle around AI has done more damage to public debate than the technology has done to employment. The UK's official data tell the same story. The claim that AI is already causing mass job losses does not survive contact with the numbers.

Yet the dismissal of risk is just as misplaced as the prediction of doom. The danger is not sudden collapse but slower erosion: weaker wage growth in routine cognitive jobs, a widening gap between high-skill tech roles and everything else, and a political backlash that could take the form of protectionism or subsidy-driven industrial policy. The Big Four accountancy firms, for example, have cut graduate recruitment by between 6% and 29% in a single year, partly because AI can already perform much of the routine data work that once required human analysts. If entry-level knowledge work continues to shrink, the pipeline for mid-career roles will dry up. That is a problem that will not show up in aggregate unemployment figures.

The better answer is not to resist AI adoption, which would be a fool's errand in a globally connected economy, nor to leave adaptation to firms and workers on their own, which is what most governments are currently doing. The aim should be to invest in retraining programmes that target the occupations most likely to be augmented rather than replaced, to reform immigration policy so that it does not compound a demographic shortfall, and to ensure that the tax system is prepared for a shift in the distribution of income from wages to profits. If employment falls in certain sectors, income that once went to workers is likely to show up as excess returns for firms that own the models, data and distribution. A tax system built for labour income will then look increasingly obsolete.

The two-speed jobs market is real, but the metaphor is misleading. It suggests two separate races rather than one economy being reshaped by three forces that are rarely discussed together: demography, immigration and technology. AI is the most visible of the three, but it is also the most easily exaggerated. The challenge is not machines. It is whether the UK's institutions can adapt before the mismatch becomes a crisis.