Still Queuing at the Door

The story says young people will stop going to university now that the machine knows everything. The applications say otherwise. The jobs at the other end are the part that has gone.

By the 30 June UCAS deadline, 695,740 people had applied to a UK university or college, the highest number ever recorded, up 4.6% on last year. International applicants rose 7.1% to 148,350. Among UK 18-year-olds the application rate edged up to 41.4%. Then, a few weeks later, Adzuna reported 8,383 graduate vacancies in July, the lowest since it started tracking in 2016 and down 45.6% in a year. In mid-2017 the same measure was over 55,000.

I had assumed I’d be writing a different piece. The first part of this was about research leaving the university for the corporation, and the timeline before it was about where all of that work came from. The obvious follow-on, I assumed, was the story everyone tells at dinner: nobody will bother with a degree once a machine can answer any question you put to it. That story is not showing up in the data. Demand for a place is at a record. The thing that has collapsed is what waits on the other side.

The rung that went missing

Erik Brynjolfsson and his colleagues at Stanford have been tracking this in American payroll data, and they updated their findings last month. They find no evidence of widespread, economy-wide job displacement, which is worth saying plainly because plenty of people will tell you otherwise. What they do find is a gap that keeps widening: employment of 22 to 25 year olds in AI-exposed occupations now sits 19% below where it would be if it had kept pace with their less-exposed peers. Experienced workers in the same occupations show no such gap. The declines cluster in work where AI automates a task rather than assisting with it.

Put the two together and you get something quite specific. The junior rung is being removed while the ladder above it stays where it is. That isn’t a story about degrees being worthless. It’s a story about the first job being the thing that disappeared, and the first job is where most people learned to do the work.

I wrote about that in The Work Made Us, and I’ll admit I framed it then as a problem for individuals. It’s bigger than that. If the tasks a graduate used to cut their teeth on are now done by a model, and if those tasks were also how a profession reproduced itself, then in ten or fifteen years somebody has to explain where the experienced people came from. I have no answer to that. I’ve been turning it over for months and I still don’t.

What a degree was actually selling

Here’s where I think the dinner-party version goes wrong. A degree was never mainly a delivery mechanism for facts. The facts were in the library, and later they were on Wikipedia, and nobody stopped enrolling. What the university sold was something closer to three things at once: supervised practice at thinking, a community of people further along than you, and a credential that told an employer something they couldn’t easily find out for themselves.

The first two are arguably more valuable now, not less. It’s the third that’s in trouble, and not for the reason people assume.

HEPI’s 2026 student survey found that 95% of undergraduates use generative AI in some way, and 94% use it to help with assessed work. The share who put AI-generated text straight into a submitted piece has gone from 3% in 2024 to 8% in 2025 to 12% now. Nearly two-thirds say assessment has changed significantly in response. Only about a third feel their institution actively encourages them to use these tools, which tells you the guidance hasn’t caught up with the behaviour.

The credential is degrading from the inside while the market still insists on having one.

Now look at what employers actually do, as opposed to what they announce. Harvard Business School and the Burning Glass Institute studied 11,300 roles at large firms where degree requirements had been dropped. Hiring of candidates without a degree rose by 3.5 percentage points in those roles, which worked out at fewer than 1 in 700 hires across the economy. Around 45% of the companies that made the announcement changed nothing at all. The rhetoric said skills over credentials. The hiring said bring me the certificate.

So the pincer is this. Employers are still asking for the degree and mostly still mean it. Students are still queuing up to get one. And the assessment that gives the degree its meaning is being quietly hollowed out by a tool that 94% of them use on the work being assessed. The credential is degrading from the inside while the market still insists on having one. That’s a worse position than either “the degree is over” or “nothing has changed”, because it can persist for years without anyone naming it, and the cost lands on the graduate who did the work honestly and can no longer prove it.

I don’t think this is a cheating story, incidentally, and I’d resist anyone who frames it that way. A student using a model to explain a concept at eleven at night is doing what I do most working days. The problem is that we built a century of assessment on the assumption that the writing was the evidence of the thinking, and that assumption has quietly stopped holding.

Who pays for the research

This is where part one and part two meet, and it’s the bit I find most uncomfortable.

UK university research loses money. Not a little: the latest costing data shows a deficit of almost £5.49 billion on research activity in England and Northern Ireland in 2024-25, with universities recovering about 66.6% of what research actually costs, down from a peak of 77.8% in 2010-11. That gap is filled by teaching, and in particular by international students, who pay well above what it costs to teach them.

Which makes the teaching business the foundation the research sits on. And that business is shaky. The Office for Students reported in May that 35.8% of providers were in deficit in 2024-25, that non-UK postgraduate taught entrants fell 15.6% in a single year, and that the sector’s forecasts for recovery look over-optimistic. Its most severe modelled scenario has around 70% of providers in deficit by 2028-29.

So the picture from part one gets worse rather than better. The corporations took the AI researchers with money the universities can’t match. The universities fund what research they keep out of a teaching surplus that depends on international recruitment, immigration policy and, now, on the continued belief that a degree is worth paying for. Every one of those is wobbling at once. The independent voice I said we needed in the first piece is being paid for by undergraduate fees, and almost nobody outside the sector knows it.

What could make me wrong

Several things, and the first is the headline number I opened with.

Charlie Ball has picked apart the Adzuna figure, and the criticism lands. The 8,383 is an average of roles advertised per day, not the total advertised in July, and it counts only postings flagged as graduate roles or requiring a degree. In the same month the same site carried 187,672 teaching vacancies and 16,389 hospital nursing roles, and the two most common destinations for graduates a year after they finish are hospitals and primary schools. The graduate market is tight. It is probably not the cliff the front pages made of it.

The wider slump has ordinary causes too. Employer national insurance went up, the labour market has stalled, and Deloitte’s own survey of finance directors puts cost control ahead of AI as the reason entry-level hiring has been cut. AI is a strand in this, not the whole rope.

And the record applications might simply be demography. There are more 18-year-olds than there have been in decades, and the participation rate has barely moved. Perhaps the young are not making a confident bet on higher education at all. Perhaps there are just more of them, applying at the same rate as always, into a system that no longer leads anywhere in particular. I genuinely don’t know which of those two it is, and the honest answer is that we won’t know for another three or four years.

What would actually help

Fund research directly instead of laundering it through international student fees, because the current arrangement makes the UK’s science base a hostage to visa policy. Rebuild assessment around the things a model can’t sit in for: supervision, defence of your own work, practical demonstration, conversation with someone who knows the field. Say out loud to applicants what a degree now does and doesn’t promise, rather than selling employability that the vacancy data can’t support. And if you employ people, look hard at the junior roles you quietly stopped filling, because you are consuming a stock of experienced people that somebody else paid to create.

The queue outside the door is still long. What I can’t tell anyone in it, with any confidence, is what’s waiting on the other side.

A third piece looks at what the universities are building in response: measurement, evaluation and assurance, and whether the money exists to finish the job.