The Missing First Rung

1. A Harder Start

In April to June 2026, 981,000 people aged 16 to 24 in the UK were not in education, employment or training. That was 13.0 per cent of the age group, and 30,000 more people than a year earlier. ^[1]

It is a stark number, but it is broader than a count of people who wanted a job and could not find one.

The official measure, usually shortened to NEET, includes people who are unemployed and looking for work, as well as people who are economically inactive. In this quarter, those groups were estimated at 393,000 and 588,000 respectively. So the number describes a difficult threshold, rather than a single experience. It tells us that more young people were outside employment and listed education or training than a year before. It does not tell us that every one of them had applied for a job and been rejected, or that the same force had placed them there. ^[1]

For this programme, a first rung means more than a job title. It means a paid and credible way into work in which someone can practise, receive feedback, build evidence of competence and see a route onwards. The NEET measure cannot tell us how many people had access to that kind of beginning, or whether the available jobs offered it. ^[1]

That is the distinction behind the argument about Gen Z and artificial intelligence. The concern is not simply that there might be fewer junior titles. It is that the assignments through which beginners once learnt, demonstrated competence and became employable may be changing before equally credible routes are built in their place.

But a national youth statistic cannot show that AI caused this. Before looking inside AI-exposed desk work, we have to ask a more basic question: was the wider market offering fewer opportunities too?

2. The Wider Retreat

The wider market was not offering an easy start. Between May and July 2026, UK vacancies were estimated at 707,000. That was 81,000, or 10.3 per cent, below the level recorded before the pandemic, in January to March 2020. ^[2]

The vacancy total covers jobs at every level. It does not tell us how many were genuinely open to someone without experience, how many demanded a portfolio or previous commercial work, or how many contained paid learning, regular supervision and a route onwards. ^[2]

A vacancy can exist without being a first rung. But when vacancies fall across the economy, employers have more scope to ask for experience and applicants face more competition for the places that remain.

The strongest broad check on an AI-first explanation comes from a government-published analysis of LinkedIn data. A government-published analysis of LinkedIn data found that UK entry-level hiring was weak, but was falling broadly in line with the wider labour market. As of April 2026, it put the overall UK hiring rate at 14 per cent lower than a year earlier. ^[4]

That pattern does not support a national claim that AI alone uniquely broke entry into work. A broad downturn in demand remains a substantial explanation. LinkedIn does not cover every employer or every route, but its aggregate picture does not show entry-level hiring clearly separating from the wider retreat. ^{[2] [4]}

Beneath that aggregate picture, however, occupations did differ. The LinkedIn analysis found sharper entry-level falls in several knowledge-sector occupations and growth in some sales and customer-facing roles. Software engineering, graphic design, accountancy, product management, data analysis and legal assistance were among the routes with steeper entry-level falls. ^[4]

That variation changes the question. Professional, technical and creative desk work cannot stand for every British first job. But these are routes where generative AI can alter writing, code, analysis and documentation. If something distinctive is happening to beginners, this is where we would expect to find a signal.

3. A Signal, Not a Verdict

The most direct UK signal comes from a working paper examining firms and occupations between 2021 and 2025. It compared firms whose existing mix of jobs made them more or less exposed to the capabilities of large language models before ChatGPT became public.

The researchers used a difference-in-differences design. In plain terms, they compared the change over time in firms with different levels of predicted exposure, before and after November 2022, while using statistical controls for the firms themselves and for industry-wide changes over time. That is more informative than simply placing two national employment figures side by side, because it asks whether a relative difference opened up between more and less exposed firms. ^[3]

For low-seniority employment, the paper gives an estimate of minus 0.058 after November 2022, described by the authors as a marginally significant 5.8 per cent relative reduction. The estimate for senior employment was statistically insignificant. ^[3]

That figure is relative, not a claim that 5.8 per cent of every junior job in Britain vanished. It describes the difference associated with the study’s comparison of more and less exposed firms after the period began. The result matters because of where it appears in the employment structure, but its scale and meaning depend on the design that produced it. ^[3]

If the pattern reflects decisions inside workplaces, it could affect more than the number of junior titles. If that pattern reflects decisions inside workplaces, it could matter even when the headline occupation survives. A beginner does not become employable merely by receiving a job title. They need assignments that can be checked, feedback from someone more experienced, and time to turn repeated practice into judgement. A relative reduction in junior employment could therefore affect the supply of those learning opportunities, even before an entire profession appears to be disappearing. ^[3]

That is the possible missing rung: not necessarily the disappearance of a profession, but the disappearance of enough ordinary, supervised work at its lower levels for a newcomer to learn how the profession actually operates. The statistical result cannot observe that learning process, yet it gives a reason to look for it in named workplaces rather than treating the issue as an abstract forecast. ^[3]

This is the warning at the centre of the programme: experienced staff may remain while fewer beginners are given the paid practice needed to become experienced. But the study cannot show that mechanism happening inside a particular employer.

The paper’s exposure measure is an average predicted exposure of jobs, constructed from occupational tasks and the workforce a firm had before November 2022. It is not a record showing that a particular firm adopted a generative-AI tool, used it on a particular workflow, or removed a particular junior assignment. ^[3]

There are other reasons to be careful. The paper is not peer reviewed, and its low-seniority result is marginally statistically significant rather than unequivocal. Its design depends on the exposed and less-exposed groups having followed comparable paths before the change. Post-2022 shocks affecting firm growth, remote working or recruitment could still have landed differently across those groups. ^[3]

The study therefore gives us a credible lead, not a causal verdict. To understand what exposure might mean in practice, we need to move from a predicted score to an employer making actual decisions about tasks and training.

4. What the Employer Keeps

PwC UK provides that close-up. In its audit transparency report, the firm lists generative-AI uses that include drafting walkthrough documentation and producing and debugging Excel formulas. It says the work remains within the audit process, with teams reviewing the outputs and applying auditor judgement. ^[5]

These are ordinary pieces of desk work: documentation to draft, a formula to produce or debug, and an output that still needs human review. They show how tasks can change before an occupation disappears from employment statistics.

They do not show which grades previously performed each task, or how much junior work was displaced. That becomes important because PwC was also taking fewer beginners. Independent reporting said PwC’s recent intakes were a couple of hundred smaller. That might look like the missing link between changed work and fewer beginnings. But the firm’s chief people officer offered a different emphasis, saying recent UK cuts were “much more about the market in the UK” than AI. ^[6]

The employer’s explanation cannot settle the balance between demand, technology, offshoring, costs and other restructuring decisions. But the public record shows task use and a smaller entry route existing together; it does not show that one caused the other. ^{[5] [6]}

The most revealing decision concerns what PwC says it did not remove. The firm said it had chosen not to replace or offshore some routine work because junior staff needed exposure to subjective audit judgements. ^[6]

That choice recognises that the value of a beginner’s task is not confined to the immediate output. Attempting work, having it checked and learning why an apparently simple judgement is not straightforward are part of professional formation. Automation may save time on a task while also removing an occasion for supervised practice.

If an employer does not protect it, the same technology may raise the expected standard at the door. A newcomer could be asked to produce AI-assisted work, exercise judgement over it and show evidence of doing so, while receiving fewer paid chances to acquire those capacities. ^{[5] [6]}

The choice is therefore not between using AI and refusing it. It is between treating every saved minute as available for removal and treating some of the work as part of how professional judgement is formed. That choice affects whether a junior is merely expected to arrive ready, or is given a structured chance to become ready. ^{[5] [6]}

PwC does not prove that AI has reduced junior hiring. It demonstrates something more useful for understanding the problem: employers can use generative AI while deliberately deciding which learning work beginners still need. The first rung survives only if someone makes that learning function part of the design.

5. Building Another Rung

If conventional junior hiring becomes more selective, apprenticeships might appear to offer another way in. But a rising headline total does not reveal who receives that opportunity.

In England, apprenticeship starts rose overall in the first three quarters of the 2025/26 academic year. Yet under-19 starts fell by 5.4 per cent, to 63,530, while starts among people aged 25 and over rose by 17.5 per cent, to 161,670. ^[7]

These figures are provisional, cover England rather than Britain, and count starts rather than vacancies or completed programmes. They do not tell us the quality or regularity of supervision, the level of pay, the protected time available for learning, or whether a participant moved into sustained work. An apprenticeship start is evidence that a place began; it is not, by itself, evidence that a first rung carried someone upwards. ^[7]

A replacement route has to be judged by what it enables. That distinction matters because a replacement route has to do more than absorb a statistic. It has to reach people who would otherwise be shut out, provide the conditions in which competence is acquired, and leave them with something recognised by the next employer. A place that exists but is inaccessible, unsupported or disconnected from progression cannot carry the same weight. ^{[7] [8]}

One named route shows what some of those conditions look like. Deloitte describes its Audit BrightStart as a full-time, earn-while-you-learn apprenticeship for college leavers. The firm says it funds the qualification, provides paid study and exam leave, and offers mentoring from experienced team members alongside dedicated coaching. ^[8]

The route answers the demand to adapt with income, workplace practice, protected study and support from experienced colleagues. It does not require an accounting or finance background beyond Deloitte’s stated academic entry requirements. But the employer’s description does not tell us how competitive entry is, who can afford to take part or how accessible the route is to people without financial support or geographic mobility.

Deloitte also publishes one participant’s destination. Deloitte publishes an account of someone who joined directly from college at 18 and says: “Six years since joining Deloitte, I am now an Assistant Manager” and “a chartered accountant”. The account describes a four-year BrightStart pathway combining on-the-job training with study. ^[8]

That is a concrete example of paid work and formal learning leading onwards. That demonstrates possibility, not representative coverage or typical outcomes. The account was selected by the employer, and it cannot show how many entrants complete, progress or receive the same quality of support. ^[8]

So a replacement rung can be built. What has not been demonstrated is that routes with these features are broad and accessible enough to replace learning opportunities that may be narrowing elsewhere.

6. Somewhere to Adapt

Has AI broken the first rung of Britain’s career ladder? The evidence does not support a Britain-wide verdict that AI has broken the first career rung. ^{[2] [3] [4]}

The national market was already retreating, and broad entry-level hiring fell roughly alongside hiring at other levels. But inside AI-exposed professional and technical desk work, the warning is harder to dismiss. The fairest conclusion is therefore neither collapse nor reassurance. In Britain’s AI-exposed desk-work routes, generative AI may be accelerating an older squeeze by changing the routine work through which beginners learn. ^{[3] [4] [5] [6]}

Weak demand may be doing much of the damage. The direct chain from a named task being automated to fewer junior places has not been demonstrated. Yet the risk matters because tasks are not merely units of output. At the beginning of a career, they are also occasions to practise, receive correction and build evidence that someone is ready for harder work.

That is why telling new entrants to adapt is inadequate by itself. Adaptation needs somewhere to happen. That means income, time, supervised work, feedback, a recognised way to demonstrate competence and a believable route into the next job. ^{[6] [8]}

PwC’s account shows that an employer can protect some routine work because juniors need it to develop judgement. Deloitte’s route shows that paid work, study, mentoring and recognised progression can be combined. Neither establishes a national solution, but both reveal that the shape of the first rung is a decision rather than an automatic consequence of the technology.

Employers and providers can treat AI as a reason to raise the bar before offering a way over it. Or they can redesign paid learning first, preserving the practice and progression that make adaptation possible. ^{[3] [6] [8]}