In the 1950s, British surgeon Wilfred Ninewells discovered something troubling. Young doctors trained with the latest surgical techniques could perform complex procedures. But when the technology failed, when the device broke down or the unexpected happened, they were lost. They had never learned to work without it.
The technology had made the craft more accessible. And at the same time, removed its foundations.
That is exactly what is happening now with AI in organisations. And almost nobody has noticed.
The invisible shift
AI is spectacularly good at junior work. Writing first drafts, summarising data, taking meeting notes, structuring research, generating code. The work that new employees have traditionally started with.
That is efficient. It is also a problem.
Because that junior work is how people learn the craft. A marketer who has written a hundred customer-facing texts develops a feel for tone and audience. An analyst who has spent months manually cleaning data develops an intuitive sense for where the errors are. A consultant who has summarised dozens of research reports learns to distinguish signal from noise.
AI skips that phase. A junior can now be a "editor" straight away: reviewing AI output, refining it, approving it. That looks productive. It feels like a promotion. But it creates editors who have never learned to write.
Why it matters
The effect is not visible in the short term. The output looks good. The KPIs improve. The team seems more productive.
Over the longer term, something erodes. The critical eye needed to evaluate AI output is built on experience with the work itself. If you have never written a customer email, how do you know whether the AI version is good? If you have never cleaned a dataset, how do you spot an error in the AI analysis?
A generation of employees emerges who can operate the dashboard but do not understand the machine. That works fine until it does not. And with AI, it always goes wrong eventually. Especially when AI drift occurs and output quietly degrades without anyone noticing the difference.
What you can do about it
The answer lies in a deliberate choice: which work should a junior still do manually, even if AI can do it faster?
Define learning tasks. Per role, per team: which tasks are essential for building professional knowledge? Write them down. Protect them. A junior who has never manually conducted a customer study is missing a layer of understanding that no AI tool can compensate for.
Make AI a tutor, not a replacement. The most powerful use of AI for juniors is as a sparring partner. Let the junior do the work. Let AI give feedback, suggest alternatives, flag mistakes. That builds capability with AI as an accelerator of the learning process, rather than a detour around it.
Ask the meaning question. Check with the team each quarter: is the work that remains after AI involvement more or less meaningful than before? This is what we call the Meaning Test. Meaningful means: it requires judgement, it builds expertise, it gives a sense of contributing to something. If AI saves ten hours a week and those ten hours fill up with more administration, you have not improved the work. You have just moved it.
The bigger picture
Organisations that handle this well think about AI as a people question. The technology is the easy part. The hard question is: how do we make sure our people get better, not just faster? A good AI workshop starts exactly there, with the people, not the technology.
That requires an uncomfortable conversation. It requires managers to say: "I know AI can do this faster, but I want you to do it yourself first." That feels inefficient. It is an investment.
The British surgeons eventually solved it by introducing simulation training. Not to abolish the technology, but to ensure the fundamentals remained solid. The technology became more powerful. The people behind it did too.
That is the model. AI that makes work better and people who grow alongside it. Both. At the same time.