If AI Takes Over the Work of the Apprentice, Where Will Tomorrow’s Expert Come From?

The debate about artificial intelligence and employment is usually framed around jobs: how many will disappear, how many will change and how many new ones will be created. But there is another question that may prove just as important. What happens to the development of human expertise when AI performs the work through which people traditionally learned their professions?

The hidden function of junior work

Entry-level work is not merely cheap or routine labour. In many occupations it is part of the mechanism by which expertise is created. Junior employees learn by carrying out relatively straightforward tasks, seeing how experienced colleagues deal with exceptions, making mistakes, receiving feedback and gradually being trusted with more difficult decisions.

The individual task may look inefficient when compared with an AI system that can complete it more quickly. Yet the task can have a second purpose: it gives the novice experience. Remove enough of those experiences and an organisation may improve today's productivity while weakening the route by which tomorrow's experienced people are produced.

Evidence from UK employers

Research published by the Work Foundation at Lancaster University in August 2026 gives this concern a firmer empirical basis. The research analysed a nationally representative Survation survey of 1,001 UK senior business leaders.

More than one in three employers (36%) said they had reduced entry-level jobs for young people during the previous 12 months. More significantly for the AI question, 43% of employers reported investing in AI or automation that had reduced the number of entry-level roles available. Among large employers, that figure rose to 60%.

The Work Foundation is careful not to attribute the whole deterioration in the youth labour market to AI. That caution matters. Weak demand, fewer vacancies and other economic factors are also involved. But its analysis makes a particularly important observation: entry-level jobs often contain routine tasks through which young people acquire valuable work experience. If automation removes those tasks, employers may require fewer junior workers or redesign junior jobs around higher-level capabilities.

That creates a potential circular problem. Employers increasingly want recruits who already possess experience and higher-level capability, while some of the work that previously created that experience is being automated.

Atrophy is only half the problem

Much discussion of AI dependency concerns skill atrophy. A capable person delegates work to AI, practises the skill less often and gradually becomes less capable of performing it independently. That is a genuine risk, but it assumes the human capability existed in the first place.

The succession problem is different. A new entrant may never receive enough practice to develop the capability that an older generation acquired before AI. The sequence is not capability followed by atrophy. It is AI performing developmental work, reduced human practice and expertise never fully forming.

This produces three distinct capability risks:

• Atrophy — people lose capabilities they once possessed because AI now performs the work.
• Failure of capability formation — newcomers do not receive the experience through which expertise would previously have developed.
• Organisational capability loss — experienced people retire or leave and there are too few sufficiently developed successors to replace them.

The succession question

If AI takes over the work of the apprentice, where will tomorrow's expert come from?

This question does not imply that organisations should preserve unnecessary work simply to keep people busy. Nor does it argue against using AI to improve productivity. It suggests that organisations need to distinguish between work that is merely routine and work that also performs a developmental function.

Where AI removes developmental work, an alternative route to expertise may need to be designed deliberately: supervised simulations, structured case work, rotations, deliberate practice, mentoring, AI-assisted training that requires the learner to reason before seeing the answer, and periodic independent performance without AI.

Why this matters for command

The succession problem ultimately becomes a command problem. AI oversight depends on humans who can recognise implausible outputs, understand the context, challenge assumptions and intervene when circumstances fall outside the system's competence. A human can remain formally responsible while becoming progressively less capable of exercising that responsibility.

This is why organisations may eventually need something resembling Human Capability Assurance: a systematic way of identifying the human knowledge, skills and judgement that must continue to exist if AI is to remain genuinely supervised.

The central management question therefore changes. It is no longer only, “What work can AI do?” It also becomes, “What human capability does this work currently help to create, and how will we create that capability after the work has been automated?”

A productivity gain with a delayed cost

The danger is easy to miss because the benefits arrive first. AI can reduce costs, speed up routine work and allow smaller teams to produce more. Capability succession problems emerge later, perhaps years later, when experienced staff leave and organisations discover that the pipeline which once produced their replacements has narrowed.

That does not make AI adoption a mistake. It means that capability development has to become an explicit design problem rather than an accidental by-product of junior work.

For further information: ai-humancapability.com