For much of the conversation about AI and work, we’ve focused on what technology might enable us to stop doing.
The repetitive tasks. The first drafts. The research. The meeting notes. The data analysis. The admin that takes up time without necessarily making the best use of people’s skills.
And there is clearly an opportunity here. If AI can take some of that work away, people can potentially spend more time on the things humans are particularly good at: judgement, creativity, problem solving, relationships and making sense of complexity.
But there is another question we need to ask.
What if some of the work we’re removing is also the work through which people learn?
The apprenticeship void
A newly published 2026 paper describes this as the “apprenticeship void”.
The idea is relatively simple. In many knowledge-based professions, people have traditionally developed expertise by starting with more foundational tasks and gradually taking on more complex work.
Think about a junior consultant researching a client problem before contributing to the recommendation. A communications professional producing an imperfect first draft before seeing how an experienced colleague improves it. An HR practitioner analysing employee comments before drawing broader conclusions from the data.
Researching, drafting, summarising, analysing and iterating aren’t simply tasks that produce an output. They are also opportunities to practise.
Over time, people begin to recognise patterns. They understand why one approach works and another doesn’t. They encounter the awkward exceptions that aren’t covered in the textbook. They make mistakes, receive feedback and gradually develop something much harder to teach explicitly: judgement.
Increasingly, these are precisely the tasks AI can perform for us.
The paper argues that this creates a potential gap between performing well with AI and developing genuine expertise. Someone may be able to produce high-quality work much earlier in their career, while having fewer opportunities to develop the underlying capability that would once have come from producing it themselves.
Efficiency and development aren’t always the same thing
This creates an interesting challenge for organisations.
Imagine AI can produce a competent first draft in two minutes, while asking a junior employee to produce one takes two hours. From a productivity perspective, the decision looks obvious.
But viewed through a development lens, the calculation becomes more complicated.
Those two hours weren’t necessarily wasted. They may have been part of how that person learned to structure an argument, distinguish useful information from noise, understand the subject and recognise what good looks like.
That doesn’t mean we should preserve inefficient work simply because “that’s how people learn”. AI can remove plenty of low-value activity, and few people will mourn spending less time formatting documents or manually processing information.
The opportunity is to be more deliberate about distinguishing between work that is simply inefficient and work that carries hidden developmental value.
If we automate the latter, we need to think about where that learning happens instead.
Human skills need somewhere to develop too
This isn’t only about technical expertise.
Recent reporting from the professional services sector suggests firms including EY are thinking about how junior employees develop interpersonal skills as AI takes on more technical and routine work. There is renewed emphasis on face-to-face interaction, partly because consulting has traditionally relied on an informal apprenticeship model in which less experienced employees learn by observing and working alongside more experienced colleagues.
It highlights a wider issue. We often talk about the human capabilities that will become more valuable alongside AI: empathy, curiosity, critical thinking, communication, judgement and creativity.
But people don’t magically acquire those capabilities simply because AI has made them more important. They need opportunities to practise them.
Perhaps AI adoption needs a learning lens
This is where AI becomes a work design question rather than simply a technology question.
When organisations redesign roles and workflows around AI, perhaps one of the questions they should ask is not just “Could AI do this?”, but “What does doing this currently teach someone?”
If a task disappears, where will that experience now come from? Could junior employees work alongside AI rather than simply accepting its output? Could they critique drafts, investigate anomalies, compare approaches or explain why they agree or disagree with a recommendation? Could mentoring, observation and feedback become more deliberately built into the work?
In other words, we don’t necessarily need to protect the old apprenticeship. But we may need to design a new one.
Because the risk isn’t simply that AI replaces some junior work. It’s that we become so good at optimising today’s work that we inadvertently remove the experiences needed to develop tomorrow’s experts.
At People Lab, we help organisations take a human-centred approach to designing work and employee experience. If you’re exploring how AI is changing roles, skills or ways of working, we’d love to help you think beyond adoption and redesign the experience around the humans working alongside it. Get in touch to find out more.


