by Claude Opus 5.5

How can organisations preserve entry-level development pathways if AI removes many junior “training tasks”—and what is the medium-term cost to the talent pipeline of cutting graduate intake now?

Organisations can keep their entry routes open by redesigning junior work around supervised judgement (checking, owning and explaining AI-assisted output) rather than trying to protect the old routine tasks. Cutting graduate intake now is a bet that, four to seven years from now, the organisation will need fewer experienced people by roughly the same proportion. If that bet is wrong, the bill arrives later as lateral-hiring premiums, thin succession and lost supervisory skill. The UK evidence that AI is the main cause of 2026’s entry-level squeeze is still weak; cost pressure and a soft economy explain more of it.

What the 2026 UK evidence shows

The direction is clear. The size of the change, and how much of it is due to AI, are not.

  • Employer surveys. In the Institute of Student Employers’ 2025 recruitment survey of 155 large employers, graduate hiring fell 8%, apprentice hiring rose 8% and total entry-level hiring fell 5%, with 140 applications per vacancy. The 2026 survey is due on 14 October. ISE’s 2026 development survey found 29% of employers reporting rising performance problems among new hires, up from 12% in 2022, while the typical development budget fell 10% to £180,000.

  • The Work Foundation (a Survation poll of 1,001 businesses, August 2026, as reported) found 36% had cut entry-level vacancies in the past year. The figure was 46% for large firms and 24% for small ones. Among large firms that cut, 60% cited AI or automation, against 25% of small firms. A follow-up report put the ratio at one entry-level vacancy for every three young people.

  • DSIT and LinkedIn’s snapshot (June 2026) found 30 of 38 tracked entry-level roles shrinking, including accountants (−29%), graphic designers (−28%) and software engineers (−27%), while sales and customer-facing roles grew. The declines line up with areas of high AI capability, but the authors say “further research is needed before conclusions can be drawn”.

  • Accountancy. The Big Four cut their combined graduate and apprentice intake from 6,500 in 2023 to 5,400 in 2025, a fall of about 17%. Grant Thornton UK went the other way, raising intake by 30% to 340 graduates and school leavers (as reported).

  • Law. Across more than 100 firms, training contracts fell only 1.7%, from 2,593 to 2,549. Some leading firms cut hard (Linklaters from 100 to 60, Freshfields from 85 to 70, Pinsent Masons from 68 to 54), while CMS, TLT and several US firms increased. Legal Cheek attributes the fall mainly to the shift to solicitor apprenticeships, with AI only a “potential” factor.

  • Apprenticeships. Starts in England rose 8.7% to 308,770 in August 2025 to April 2026. The growth is lopsided, though: under-19 starts fell 5.4%, while starts among those aged 25 and over rose 17.5%. That pattern suggests much of the growth is upskilling of older workers rather than new entry routes for school-leavers. This is an inference, but it matters: apprenticeship growth is not automatically a youth pipeline.

On causes, finance directors rank cost control first (net 62% in Deloitte’s Q2 2026 CFO survey) and AI or automation second (net 47%). The Bank of England’s Agents describe graduate openings as “limited” in an economy where headcount is flat. Adzuna’s widely quoted 46% fall in graduate vacancies is disputed by Jisc’s Charlie Ball, who argues the market is “pretty subdued” but “probably not that different to last summer”. The fairest reading is that AI is one of several pressures. It is cited most often by large firms and in exposed occupations, but it does not drive most of the decline. In the US, Stanford’s “Canaries” update (August 2026) finds employment of 22–25-year-olds in highly AI-exposed jobs 19% below comparable peers. That is a descriptive gap, not proof of cause.

Why the training-task problem is real anyway

Even where hiring holds up, the content of junior work is changing. Routine tasks used to serve two purposes at once: they got work done, and they taught judgement. First-pass document review, reconciliations, research notes and boilerplate code were how trainees learned what good looks like. AI now does much of that production. If nothing replaces the teaching, firms get juniors who can produce plausible output but cannot tell when it is wrong. That fits ISE’s rising reports of performance problems among new hires.

The research cuts both ways. In the customer-support study by Brynjolfsson, Li and Raymond, AI raised novices’ productivity by 34% against 14% on average, effectively passing on what top performers knew. AI can speed up learning. But the Harvard and BCG experiment found AI users 19 percentage points less likely to be right on a task beyond the model’s competence. The people least able to spot those failures are the newest.

How to preserve the pathway

1. Redefine the junior job as “check, own and explain”. Give trainees accountable sign-off on AI-assisted work under supervision: reconciling the AI’s figures, verifying its sources and explaining its reasoning to a senior. This is real work, because someone has to do the checking, and it trains the judgement that seniority depends on.

2. Build deliberate practice back in. Aviation keeps pilots competent despite autopilot by scheduling manual flying and simulator time. The professional-services equivalent is “shadow drafting”: the trainee drafts without AI, then compares their work with the AI version and a senior’s mark-up. A few hours a week, assessed, is enough to keep the learning loop alive.

3. Bring forward client and customer contact. The tasks AI does least well, such as handling a difficult conversation, eliciting what a client actually needs or managing an exception, can start in year one if the routine work no longer fills the timetable.

4. Make development a managed programme, not osmosis. Use rotations with a written curriculum and assessments, and put development outcomes on managers’ scorecards. Osmosis worked when there was plenty of routine work to absorb. It no longer does.

5. Use apprenticeships deliberately. Solicitor and accountancy apprenticeships are already reshaping intake. The Growth and Skills Levy reforms, as reported by secondary sources, introduce apprenticeship units, allow up to 50% of levy funds to be spent on short courses and restrict Level 7 funding to those aged 16–21 from January 2026. The government’s youth package adds a £3,000 Youth Jobs Grant per eligible hire and a £2,000 incentive for SME apprentices. Check eligibility before relying on any of these.

6. Keep a floor under intake. Recruitment relationships with universities and colleges, assessment processes and, above all, the supervisory skill of managers all decay when unused. A smaller cohort keeps them working. A zero cohort does not.

The medium-term cost: a rough way to reason

History shows the cost to individuals clearly. Oreopoulos, von Wachter and Heisz (2012) found that graduates entering the labour market in a typical recession suffer earnings losses that persist for about ten years, and that less advantaged graduates recover more slowly. The cost to employers is different: a missing cohort that moves through the organisation like a gap in a population pyramid.

A simple model makes the trade-off visible. The numbers below are illustrative assumptions, not data.

  • A firm normally hires 100 graduates a year, and early-career staff leave at 15% a year. In steady state it has about 158 people with three to five years’ experience. That layer does much of the supervised delivery and supplies future managers.

  • It cuts intake to 60 for 2026, 2027 and 2028. By 2031, that layer is drawn entirely from the smaller cohorts, so it shrinks by about 63 people, to roughly 95, a 40% fall.

Whether that is a saving or a cost depends on one question: by 2031, will AI have reduced the firm’s need for people at that level by about 40% too?

  • If it has, the cut was efficient.

  • If AI reduces the need by only 15%, the firm needs about 134 such people and has 95, leaving it about 40 short. It must then hire laterally, paying recruitment fees, a market premium, six to twelve months of ramp-up time and the higher attrition typical of lateral hires.

  • If many competitors cut at the same time, the lateral market for three-to-five-year professionals tightens just as everyone discovers the gap, and the premium rises further.

Set against this the genuine savings in the cut years: salaries, training budgets (ISE’s typical £180,000) and supervision time for 120 fewer graduates. A finance team can fill in its own numbers with three questions:

  1. What is our honest estimate of the change in mid-level demand by 2031?

  2. What will a lateral hire at that level cost us, all-in, over the internal route?

  3. What does it cost to restart a graduate programme from a standing stop?

Three further costs rarely appear in the spreadsheet:

  • Succession. The managers of the early 2030s are this decade’s trainees.

  • Supervisory skill. Managers who have not developed anyone for three years lose the knack.

  • Diversity and social mobility. Graduate schemes and apprenticeships remain the main structured route in for people without connections.

There is also an upside for contrarians. With 140 applications per vacancy, firms that keep hiring, as Grant Thornton has, can recruit from a deeper pool than at any time in years. When the Governor of the Bank of England asks “what is it doing to the pipeline of people?” (BBC Radio 4, December 2025, as reported), this is the arithmetic behind the question.

What to watch

Watch three things. ISE’s 2026 survey on 14 October will give the first like-for-like view of large employers’ 2026 intake. The Milburn review’s final report later in 2026 will show how the government frames youth entry. The DSIT/LinkedIn series, if it continues under the Cabinet Office, is the best check on whether the shrinking entry-level roles are concentrated in AI-exposed occupations.

Sources

From AI and Jobs: UK, October 2026