Your senior team of 2035
is being hired now

The reassuring version of the AI jobs story may also contain the management problem that matters most.

So far, the better US evidence does not show a broad purge of established workers. Anthropic’s March 2026 labour-market report found no systematic increase in unemployment among workers in highly AI-exposed occupations.¹ Stanford’s August revision likewise found no widespread, economy-wide displacement.² Yale’s Budget Lab has been more cautious still: its 14 October update found no clear large-scale, economy-wide disruption across occupational churn, employment and wages, although it flagged a volatile rise in unemployment among AI-exposed occupations that was not corroborated by its other measures.⁴

That is genuinely reassuring. It is also a reason to look somewhere other than redundancy statistics.

A labour market can adjust without large numbers of existing employees losing their jobs. Organisations can simply hire fewer people at the bottom. The saving appears immediately. The consequence may not become visible until the missing cohort should have become the experienced workforce.

The quieter signal is appearing at the entry point

Anthropic’s most interesting finding was not a surge in unemployment but a tentative change in the job-finding rate among younger workers. Using US Current Population Survey data, the researchers estimated an approximately 14 per cent decline in the job-finding rate for workers aged 22–25 entering highly AI-exposed occupations relative to 2022. The result was only just statistically significant, and Anthropic found no comparable decrease among workers over 25.¹

Taken alone, that would be a useful watchpoint rather than a strong conclusion. But later evidence has made the pattern harder to dismiss.

A US Census Bureau working paper using matched employer-employee administrative data found an immediate and persistent fall in early-career hiring in the most AI-exposed industry-state groups after ChatGPT’s introduction. Regression-adjusted employment among workers aged 22–24 in the most exposed quintile fell 12 per cent over the following ten quarters, with reduced hiring identified as the main mechanism.³

Stanford’s Digital Economy Lab then extended its large ADP payroll analysis through June 2026. It found that employment among 22–25-year-olds in highly exposed occupations was about 19 per cent below where it would have been had it kept pace with similarly aged workers in less-exposed occupations. Again, the adjustment appeared to operate mainly through lower hiring rather than higher separations.²

None of this settles causality. Stanford explicitly describes the patterns as descriptive; some differences pre-date generative AI, and education and other labour-market forces complicate the comparison. Yale’s work remains a useful counterweight because its October update still finds no clear large-scale, economy-wide disruption across occupational churn, employment and wages, while flagging a volatile rise in unemployment among AI-exposed occupations that is not corroborated by its other measures. The sensible conclusion is not that AI has been proved to be causing a junior-employment collapse. It is that entry-level hiring has become a credible management signal worth measuring inside your own organisation.²⁴

On a narrow screen, each period is shown as a separate card.

Reading the figure: Anthropic reports an approximately 14 per cent pooled decline in the job-finding rate for workers aged 22–25 entering highly exposed occupations. The estimate was only just statistically significant, was based on survey transitions and is not causal proof. No comparable decrease was found for workers older than 25.

Think like a demographer, not only a cost controller

An organisation has something resembling a population pyramid. People enter in cohorts, accumulate experience, move through roles, leave, return, specialise and eventually form the pool from which more senior positions can be filled.

If AI allows a department to recruit 12 graduates where it previously recruited 20, the immediate arithmetic may be excellent. Output per employee rises. Salary cost falls. Nobody has been made redundant. A finance dashboard can register an uncomplicated productivity gain.

But the eight people not hired are also absent from the organisation’s future experience pipeline. Repeat the decision for several years and a narrower cohort moves through the structure. Five or ten years later, management may discover that the internal pool of experienced specialists, supervisors and future leaders is thinner than expected.

The demographic analogy is not exact. Companies can hire laterally, retrain, accelerate promotion, redesign work and recruit internationally. A missing workforce cohort is therefore not as fixed as a missing birth cohort. But the constraint is still baked into the choices available later: if fewer people acquired the relevant experience, they cannot be conjured into existence when demand returns.

On a narrow screen, each exposure group is shown as a separate card.

Reading the figure: Stanford’s August 2026 revision reports a 19 per cent relative shortfall for the more exposed group against the pace of less-exposed peers. Stanford stresses that the patterns are descriptive rather than definitive causal estimates.

The pipeline is about more than headcount

Stanford’s latest revision adds a useful clue. Younger employment has weakened more in occupations where knowledge is readily codified, while experienced workers in more tacit-knowledge-intensive work have held up better. AI can therefore remove routine junior tasks without necessarily removing the later need for human judgement.²

Andrew’s Intel has examined that learning problem separately in the approved corpus analysis AI productivity: the hidden capability tax.⁶ That analysis develops the risk that an organisation can remove the work while still needing the judgement, context and verification capability that the work used to develop. The full argument does not need repeating here. For workforce planning, the narrower point is enough: a smaller entry cohort may also receive a thinner apprenticeship unless development is deliberately redesigned.

The apprenticeship trap is an old coordination problem

There is a useful precedent in employer-funded training. British research has found that employer-provided training is less frequent in denser labour markets, where potential poaching and turnover are greater. The logic is familiar: one firm pays to train a worker; another firm can later hire the trained worker and capture part of the benefit.⁵

At the level of the individual organisation, reducing training can therefore be rational. If competitors are willing to supply experienced people to the market, why should you carry all of the development cost yourself? The problem appears only when enough firms make the same sensible calculation.

Then the shared pool of skilled people becomes smaller. Recruitment gets harder. Wages rise. Firms bid against one another for the same experienced workers. Organisations that continued to train can become attractive hunting grounds for competitors. Their retention costs rise, which can make training look still less attractive.

AI can create a white-collar version of the same loop. If every organisation plans to hire fewer juniors today and buy experience from the market later, the market later may contain less experience to buy.

On a narrow screen, each stage is shown as a separate card.

Boundary: This is an illustrative mechanism, not a forecast. It shows how rational short-term choices can make the shared future talent pool thinner.

“We can hire experienced people later” is not a free option

External recruitment is often treated as a release valve in workforce planning. It is real, but it is not free. It transfers the cost and risk of development to somebody else.

If an industry collectively reduces its entry-level intake, the price of imported skills should rise. Scarce mid-career people gain bargaining power. Employers must offer more to recruit them and more to retain the people they have already developed. The conscientious organisation that maintains its own pipeline may still lose trained staff into the tighter market precisely because they have become more valuable.

That creates a second-order risk. A firm keeps training, loses more trained people to competitors, sees the private return on training deteriorate and begins to ask why it should continue. The individually rational response compounds the collective shortage.

This is why entry-level hiring should not be evaluated only as current labour demand. Some of it is investment in future organisational capability. The accounting line is payroll; the economic function may be closer to replenishing productive capital.

Scarcity can climb the salary ladder

If experienced mid-career people become scarce, the price of importing them rises. That pressure need not remain neatly contained at the scarce grade. When the market rate for a specialist moves closer to the pay of the person managing them, organisations must either accept compressed differentials or lift salaries, retention packages and promotion expectations further up the structure.

That is a management implication, not a measured outcome of the current labour-market studies. But it shows why a junior-payroll saving can return later in a different form. The organisation may need fewer people overall and still find the remaining human workforce considerably more expensive to recruit and retain.

The serious counterargument: perhaps the whole pyramid shrinks

AI may not merely hollow out the base. As systems become useful further up the employment value chain, organisations may need fewer people at intermediate and even senior levels. A firm that once needed a broad junior intake, a large specialist layer and many managers may eventually need a much slimmer structure at every level. If so, lower entry-level hiring today could be prudent anticipation rather than underinvestment.

That possibility is exactly why the answer is not “keep hiring juniors”. The planning question is: what shape of human organisation do we expect to need in five to ten years, and what intake and development system will produce it? If AI removes labour roughly proportionately across levels, the cohort risk is smaller. If it removes routine junior work faster than it removes demand for experienced judgement, the organisation becomes narrower at the base and more dependent on a scarce upper cohort. Either way, management needs an explicit model rather than an assumption.

What managers should do with the signal

This is not an argument for preserving junior headcount regardless of need. Nor is it a prediction that every organisation cutting graduate recruitment will face a leadership crisis in 2035. The evidence on AI and employment is still early, and the right workforce shape will differ by function and sector.

It is an argument for widening the business case.

When AI reduces the amount of entry-level labour required, management should model at least three things at once: the immediate productivity gain, the future cohort effect, and the plausible shape of the human organisation if AI continues moving up the value chain. That means asking where senior capability currently comes from, which experiences create it, how quickly people leave, how much future demand can be met internally, what happens to pay and retention if mid-career skills tighten, and whether the future organisation is likely to shrink proportionately or become unusually thin at the base.

The result may still be fewer entry-level hires—perhaps materially fewer. But it should be a planned reduction accompanied by a redesigned development path and an explicit view of the future workforce pyramid, not a silent assumption that tomorrow’s experienced people will appear in the labour market at yesterday’s price when needed.

The Workforce Cohort Test

Before booking the full saving from lower junior recruitment, ask:

On a narrow screen, each question is shown as a separate card.

Reassuring evidence is not permission to stop planning

For established workers, the current evidence is more reassuring than many headlines suggest. There is still no clear sign of economy-wide AI displacement, and experienced workers appear relatively resilient in several of the datasets now being watched.

For managers, however, that reassurance changes the question rather than ending it. The first important workforce effect may arrive through vacancies that are never opened, graduate classes that are quietly smaller and entry-level tasks that disappear from the development path.

Those choices can be perfectly rational this year and still create a more expensive organisation later. The risk is not that management fails to preserve today’s pyramid. It is that management takes the quick win without deciding what human pyramid it is trying to build next.

Your senior team of 2035 is not an abstract future resource. A meaningful part of it is being hired, trained and retained now—even if the team itself will eventually be smaller. AI can make senior expertise more valuable while weakening the system that produces it. The management task is to take the productivity gain without allowing the future capability pipeline to become an accidental casualty of today’s optimisation.