In Article 3, I considered a hypothetical 20-person equipment servicing company using AI to help prepare engineer service reports. Suppose that approach works. The reports are completed more quickly, checking is manageable and customers receive the information sooner. A sensible improvement, perhaps. But it raises a less immediate question: what happens to the junior engineer who would previously have learned by preparing those reports?

The attraction of automating routine work is obvious. However, some of that work has traditionally been where employees learned the detail of a business, recognised mistakes and gradually developed the judgement required for more responsible roles. Removing the task without replacing the learning could create a problem that does not appear in the first year’s productivity figures.

What the Anthropic figures actually show

Anthropic’s October 2026 study found that the junior share of technical hires fell from 27% in early 2025 to 9% by mid-2026. Junior employees also declined as a proportion of its technical workforce, from 21% to 14%. The company was still growing and technical hiring continued to increase in absolute terms. These are changes in the composition of recruitment, not evidence that Anthropic stopped hiring junior staff. [1]

The study also described managers placing greater weight on autonomy, judgement and the ability to work effectively with AI. Employees were increasingly directing and checking agent output rather than carrying out every implementation task themselves. Anthropic is an unusually AI-intensive organisation, so its experience cannot simply be applied to UK businesses. Nevertheless, I think there is a question worth considering: how does somebody acquire expert judgement if much of the work that once developed it is now performed by an AI system? [1]

That is not an argument against AI. It is an argument for treating workforce development as part of the adoption strategy, rather than assuming that the traditional route from junior to experienced employee will look after itself.

The UK picture is less straightforward

A June 2026 analysis by the Department for Science, Innovation and Technology and LinkedIn found that UK hiring was 14% lower year-on-year in April 2026. Entry-level hiring was declining broadly in line with the wider market, although particular junior information-processing roles were weaker: software engineer hiring was down 27%, for example. The researchers explicitly cautioned that this does not establish AI as the cause; economic conditions and changing employer requirements also matter. [2]

The ONS reported in October 2026 that, among UK businesses employing at least ten people and already using AI, 6% said adoption had reduced their overall headcount. Some 38% identified administrative and clerical roles as those most affected. These are self-reported estimates, not evidence of a general collapse in employment. They also say relatively little about how opportunities for new entrants may change before total headcount does. [3]

There is a related finding from the Institute of Student Employers. In evidence to Parliament, it reported that 43% of employers surveyed said early-career roles had evolved with AI without being formally redesigned; only 1% reported a formal redesign. This comes from participating graduate and early-career employers, rather than a representative survey of every UK business, but the distinction is revealing. Work can change faster than the development arrangements around it. [4]

The Development Test

Could a new employee explain why the AI output is right, identify when it is wrong and know when to involve somebody more experienced? If not, the business may have removed the task without replacing the learning.

Producing an answer is not the same as understanding it

A junior engineer might learn a great deal by preparing a service report: identifying the relevant equipment details, understanding what was done, spotting inconsistencies and asking an experienced colleague why a conclusion is justified. Some of the work is repetitive. That does not mean it is without value as a learning exercise.

If AI prepares a convincing report in seconds, the junior employee can appear productive without necessarily understanding the technical reasoning. Worse, a polished but incorrect answer may be harder to challenge without the underlying knowledge. The same issue could arise in software development, finance, legal support or customer service. The risk is not that all routine work must be preserved. It is that competence may be assumed because the output looks competent.

Redesign the junior role, not just the workflow

Returning to the hypothetical servicing business, I would not insist that a new engineer writes every report from scratch indefinitely. Instead, I would deliberately retain a route to technical competence alongside the AI-assisted process.

At first, a new starter could prepare selected reports independently and compare the work with an AI draft, discussing differences with a mentor. Later, the employee might check AI-assisted reports and explain why particular details have been accepted or challenged. Only when competence is demonstrated would routine review become more independent, with clear escalation for unfamiliar or higher-risk cases.

That approach also depends on experienced employees having time to teach. A senior engineer who is expected to supervise AI output, meet customers and mentor colleagues may simply acquire three competing demands on the same working day. Supervision and coaching therefore need an explicit owner and sufficient time, rather than being left to goodwill.

There is room for AI within the training itself: preparing practice scenarios, explaining alternative approaches or helping employees explore unfamiliar concepts. But the employee must still be able to test the answer against trusted information and explain the reasoning. The objective is capability, not familiarity with a particular tool.

A proportionate approach for smaller companies

For a business with perhaps ten or twenty employees, a large graduate scheme is unlikely to be realistic. The practical starting point is to identify which junior tasks are being shortened or removed, what those tasks used to teach, and where that learning will now take place.

I would keep the initial plan simple: appoint someone accountable for development; identify a small number of essential skills and real examples of competent work; set aside supervised practice; and review progress at agreed points. Where appropriate, include apprentices, career changers and existing employees moving into new responsibilities. Entry-level capability is not just about age.

This is consistent with Skills England’s July 2026 employer guidance, which recommends practical AI training connected to real work, including feedback, repeat practice and the ability to use tools safely and critically. The guidance is intended to support employers with limited capacity as well as larger organisations. It is a useful reminder that tool access and workforce competence are not the same thing. [5]

Tomorrow’s expertise needs a starting point

There may be a sound commercial case for automating tasks previously undertaken by junior employees. There may also be a sound case for changing the skills expected of new recruits. I would be concerned, however, if the result were to recruit only people who already possess the judgement that the business has stopped giving newcomers a chance to develop.

The UK government has recognised the issue through its Early Careers Jobs Alliance, established to investigate how entry-level work is changing and develop practical approaches to preserving progression routes. That is encouraging, but smaller employers do not need to wait for national guidance to review their own arrangements. [6]

My view is that a responsible AI strategy should ask two questions together: which work can be improved today, and which capabilities must still be developed for tomorrow? The first may produce a quick efficiency gain. Neglecting the second could make that gain rather more expensive in the longer term. In Article 6, I will turn to another practical requirement: managing the risks of AI already being used inside a business.

Richard James Bowles is Founder & Principal of Richard James Advisory Ltd, providing independent senior technology, transformation and board-level advice to founder-led and growing businesses.

Sources and methodological notes

[1] Anthropic Institute, How employment is changing at Anthropic (16 October 2026). Internal company data; unusual AI-intensive employer; junior percentages concern hiring composition. Source

[2] DSIT and LinkedIn, A snapshot of entry-level hiring in the UK (8 June 2026). LinkedIn hiring data; occupational differences are not causal evidence of AI displacement. Source

[3] ONS, Business insights and impact on the UK economy: 8 October 2026. September BICS responses from AI-using firms; self-reported employment effects; statistics in development. Source

[4] Institute of Student Employers, Written evidence AIB0029 to the Business and Trade Committee (submitted April 2026). ISE Development Survey 2026; employer-member sample. Source

[5] Department for Work and Pensions and Skills England, Employer guide: What works for AI upskilling in the UK (updated 27 July 2026). Practical training guidance; applies to England. Source

[6] DSIT and DWP, Entry-level jobs support, AI bootcamps and tech training (6 June 2026). Announcement of the Early Careers Jobs Alliance. Source