Artificial intelligence is not killing jobs in Britain. It is killing the career ladder

THE PROMISE of artificial intelligence was once told as a story of abundance: machines would do the drudgery, and humans would do the interesting work. A new story is emerging from the data. AI is indeed lifting humans into more interesting work-but it is also removing the stepping-stones that get them there. The result is not mass unemployment. It is a broken career ladder.

PwC's 2026 AI Jobs Barometer, based on more than a billion job adverts globally, finds that specialist AI job postings in the UK surged by 61% over the year, from 112,000 to 180,000. The average wage premium for workers with AI skills tripled to 34.2%, up from 11% in 2024. Specialist AI roles now account for 2.2% of the overall UK job market, up from 1.3%-even as total vacancies across the economy fell by 6.6%. The rebound is being driven not by people who build AI models but by those who apply them: so-called AI user roles grew by 66%, compared with 22% for developer positions. The experimentation phase is over, as PwC puts it. The scaling phase is a different sort of game.

The barometer describes a two-track labour market. Jobs that AI "professionalises"-roles where the technology removes routine tasks and frees workers for higher-value judgement, creativity and leadership-are growing twice as fast as jobs it "democratises", which are simplified but expanding more slowly. Professionalised roles have seen 42% faster wage growth since 2021. Crucially, the most AI-exposed junior positions are now seven times more likely to demand traditionally senior skills such as leadership and strategic thinking, compared with the least exposed junior roles. Meanwhile, overall entry-level postings in highly AI-exposed sectors have flatlined. The traditional career ladder is compressing: the ground floor is disappearing, while the top floor gets a pay rise.

To be sure, the headline danger-widespread job destruction-has not materialised. A study by British Progress, a think-tank, examined Annual Population Survey data covering 412 UK occupations and found no evidence that AI has replaced jobs at scale. The UK government's own assessment, published in January, echoed the caution: while around 70% of British workers are in occupations that AI could potentially affect, exposure is not the same as displacement. The Office for National Statistics reported in July that the unemployment rate was 4.9%, up from 4.7% a year earlier, and that vacancies fell to 712,000. These figures tell a story of a tight but not collapsing labour market.

The deeper problem is not displacement. It is selection. AI has made the entry-level rung economically redundant for many employers. Why pay a graduate to write boilerplate, unit tests, or simple database queries when an AI coding agent can produce them instantly for the cost of a cloud subscription? Harvey Nash, a recruitment firm, describes the current UK software engineering market as "increasingly senior-driven", with employers expecting even entry-level candidates to hit the ground running with broad technical expertise. IT Jobs Watch data shows that while the median salary for a senior software engineer held steady at £75,000, the wider market compressed at the high end, with the 90th percentile dropping from £130,000 to £100,000. Employers are hiring more selectively, not less.

The incentive structure is clear. AI raises the productivity of senior workers who can evaluate, orchestrate and integrate automated output. It lowers the marginal cost of the tasks that juniors once performed. The former becomes more valuable; the latter loses its economic rationale. This is not a bug in the system. It is the system working exactly as its architects intend.

The trouble is that this dynamic creates two second-order problems that will not resolve themselves. The first is a pipeline crisis. For decades, companies invested in juniors with the expectation that two or three years of grunt work would produce productive mid-level engineers. That deal is breaking. Fewer juniors are being hired, so there will be fewer seniors in five years. The senior shortage is already acute; the problem will compound. Senior developers who work effectively with AI tools can often deliver the same output that previously required additional junior support. But seniors are not infinitely elastic. They face a "review tax", spending hours auditing AI-generated code for subtle errors, and a "delegation vacuum", with no lower-risk tasks to hand off. Burnout is the hidden cost.

The second problem is distribution. When the bottom of the ladder narrows and the top gets a premium, wages polarise. PwC's barometer shows that the AI wage premium peaks at 64% in consumer markets. Senior roles in regulated industries such as finance, gambling and gaming command 15% to 30% premiums over generalist roles. The result is a labour market that increasingly rewards orchestration and penalises routine execution-the very combination that entrenches inequality along the lines of who gets to become a senior in the first place.

It is tempting to think the answer lies in reskilling. The government's assessment calls for "proactive management of transition risks" and PwC urges firms to "reinvent early career pathways". These are necessary measures, but they address symptoms rather than structure. Reskilling programmes that teach people to prompt AI tools do not restore the economic reason for hiring them. Mentorship programmes are a noble idea when the people meant to be mentored are no longer on the payroll.

The better answer is to rebuild the training ground through different incentives. One option is tax relief or wage subsidies for firms that hire and train entry-level talent in AI-exposed sectors, structured not as a general subsidy but as a conditional one: firms receive support in proportion to the number of early-career hires and the retention rate after two years. A second is to encourage sector bodies, professional associations and universities to co-design apprenticeship and sandwich-year programmes that embed students in real projects alongside AI tools, so that graduates arrive with the hybrid skills employers now demand. The idea is not to preserve old ways of working but to accelerate the development of the new ones.

Another, harder, lever is regulatory. If AI adoption concentrates economic rents in the hands of workers and firms that own the models, data and orchestration tools, a tax system built for labour income will look increasingly obsolete. The government's assessment acknowledges the issue but stops short of recommending a structural rethink. Better to anticipate the concentration and design a response than to wait for it to show up in inequality statistics.

The Luddites were wrong about machines in the long run, but they were right about the pain of transition. Britain's labour market is not facing an apocalypse. It is facing a rerouting. The question is whether the reroute leads to a wider, higher-stakes economy or a narrower, more exclusive one. The choice is not between AI and jobs. It is between an economy that invests in the ladder or one that removes the rungs and expects everyone to jump.

That bargain is breaking. Policy ought to mend it before the pipeline runs dry.