A curious reversal is taking place in the journey into adulthood. Young people leave home at eighteen, spend three or four years at university becoming educated and independent, then graduate and move back into the bedroom they thought they had left behind. Housing costs explain part of it, but something else is becoming harder to ignore: for many graduates, the first proper job is becoming extraordinarily difficult to get.
The numbers are uncomfortable. The Institute of Student Employers says major employers are now receiving about 140 applications for each graduate vacancy, compared with 38 two decades ago. The ONS reports that 28.7 per cent of 20 to 34-year-olds were living with their parents in 2025, up from 25.4 per cent a decade earlier. Those figures do not prove that graduates are moving home because AI took their jobs. They do show a labour market already under strain just as AI begins to change the economics of entry-level work. (ISCE)
You can see the human version in young people applying for dozens of jobs, sometimes more than a hundred, and hearing little or nothing back. They were told to work hard, go to university and acquire useful skills. Many did exactly that. They are now discovering that the first rung of the career ladder has become crowded just as employers are asking whether they need quite so many rungs.
The attraction for companies is obvious. If AI can draft contracts, write routine software, analyse accounts, summarise research and prepare presentations, firms will need fewer people doing work traditionally given to trainees and graduates. Each decision may be rational. The problem is that junior work has always done two jobs at once: it produced useful output and, almost accidentally, produced experts.
The senior engineer who knows that an architecture feels wrong has usually spent years seeing smaller systems go wrong. The lawyer who notices the dangerous clause has read thousands of harmless ones. Expertise often looks like instinct from the outside, but much of it is accumulated experience compressed into judgement. AI is becoming very good at removing precisely the tasks through which that experience was acquired.
So two problems are developing together. Companies may become more efficient today while weakening the pipeline of people capable of running them tomorrow, while society accumulates well-educated young adults who cannot get the experience required to become experienced. For a while firms can recruit people trained under the old system. That is a finite resource masquerading as a labour market.
Public policy still feels oddly behind the problem. Britain has an energetic AI agenda built around compute, investment, skills and AI Growth Zones intended to accelerate data centres and supporting infrastructure. The government has begun to acknowledge the employment issue, with measures announced in 2026 around entry-level jobs, AI training and graduate traineeships. But the balance remains revealing. We are planning where the servers will live and how they will be powered with considerably more urgency than how young people will enter the professions those servers are changing. (GOV.UK)
That deserves attention because this could become a social problem before it becomes an obvious economic one. A few graduates spending an extra year at home is ordinary family life. A generation repeatedly unable to establish careers, incomes, independence and confidence is something else. If the entry point into professional work shrinks significantly, the consequences will eventually appear in housing, family formation, tax revenues and politics, by which time the problem will be harder to unwind.
The answer is not to preserve pointless junior work, nor to send every graduate on another course in how to use AI. We may need mechanisms that deliberately bridge education and employment: government-supported graduate apprenticeships, temporary wage subsidies or grants that make taking on inexperienced people attractive, and partnerships between universities, employers and professional bodies that create genuine work experience rather than another certificate.
Companies have a responsibility too. If AI removes routine production, they will have to design apprenticeship rather than assuming it happens automatically. Junior lawyers can use AI to draft but must learn to challenge the draft. Young engineers can build with coding agents but should own what they build and investigate when it fails. The aim is to automate the drudgery without automating away the experiences from which judgement develops.
For decades, junior employment, useful work and professional education were bundled together, so nobody had to think very hard about the mechanism. AI is pulling them apart while the graduate labour market is already creaking. If we treat that simply as another productivity opportunity, we may discover too late that we have built an economy with fewer entry-level jobs, fewer future experts and a large number of young people still waiting for adult working life to begin.
