What happens when AI stops teaching us how to work?

The most consequential effect of AI on organisations may not be how much work it eliminates. It may be how much learning it displaces along the way.

 

For decades, organisations have used work to accomplish two things simultaneously: produce an outcome and develop the people producing it. Junior employees learned by doing the less glamorous parts of the job. They drafted before they edited. They analysed before they advised. They solved routine problems before they were handed the reins on more complex ones.

 

Now AI is beginning to separate those two functions. It can generate the output without requiring the person deploying it to develop all the capabilities that once came from doing the work themselves.

 

The increasingly urgent question for organisations is therefore: if AI performs the tasks through which people traditionally developed expertise, where will that expertise come from?

 

Already, our Work AI Index finds that heavy AI users are automating 46% of their output, while 40% fear that their roles could eventually be eliminated. In a difficult economy, it is tempting to view those productivity gains primarily through the lens of headcount. But that misses a longer-term organisational risk.

 

The risk is not only job loss

It is the erosion of expertise inside the jobs that remain.

 

When AI takes over the tasks through which people learn, practise and demonstrate judgement, organisations can gain short-term productivity while weakening their capacity to develop expertise. A junior engineer who never has to work through the foundations of a system may produce answers faster without developing the intuition to diagnose a difficult problem. A junior lawyer who primarily reviews AI-generated analysis may become more efficient without developing the instincts to spot a subtle weakness in an argument.

 

This is not merely a hypothetical concern. In the Work AI Index, 41% of workers say they sometimes deliver AI-generated work they could not explain if asked.

 

The challenge, then, is not to preserve work for the sake of it. It is to distinguish between work that produces an outcome and work that produces capability.

 

AI does not always remove work; sometimes it shifts people from doing the work to supervising it. In our research, workers spend roughly as much time making AI output usable as they do generating useful output in the first place. We call this “botsitting”: providing context, correcting errors, verifying claims and supervising the system.

 

That time is not necessarily wasted. But it is different from doing the underlying work, and it may not provide the same opportunities to develop expertise.

 

The question leaders should therefore ask is not simply whether AI makes people faster. It is whether the new way of working makes people more capable.

 

The changing architecture of expertise

AI does not necessarily have to displace learning. It can change how learning happens.

 

Among high AI achievers in our research, 68% say they use AI as a valuable teacher, compared with just 28% of low AI achievers. That points to a broader opportunity: organisations can invest in technologies—including low- and no-code tools—not only to automate work, but to give employees more opportunities to experiment, build and develop capabilities through practice.

 

But learning to use AI is only part of the equation. One of the clearest characteristics of high AI achievers is that they know when not to use it. Among high AI achievers, 89% say they know when not to use AI, compared with 68% of low AI achievers.

 

That is not a prompting skill. It is judgement: knowing what good looks like, recognising when an answer is wrong or incomplete, and understanding when human judgement needs to take the lead.

 

The challenge is that judgement is built through experience. If AI removes too many of the tasks through which people traditionally developed expertise, organisations cannot assume that expertise will simply emerge later.

 

They will need to deliberately create opportunities for people to make decisions, wrestle with ambiguity and learn from mistakes—even when AI could produce an answer faster.

 

Redesign work for capability, not just capacity

For decades, productivity and development were bundled together. People learned because they had to do the work. AI is unbundling them.

 

That means leaders can no longer assume that the work that produces today’s results will automatically produce tomorrow’s expertise. They need to ask two questions of every major AI-enabled workflow:

 

What work should we automate because it consumes capacity without building much capability?

 

And: What work should people continue to do, at least in part, because it develops the expertise the organisation will need?

 

The answer will not always be to keep humans in the loop. Sometimes the right decision will be to automate a task completely. In other cases, human involvement may be precisely what needs to be preserved—not because the machine cannot do the task, but because doing the task is how people learn to do the next, harder one.

 

This also changes how organisations should think about AI literacy. Training people to use AI is necessary, but insufficient. The deeper capability is knowing when to trust it, when to challenge it, when to verify it and when to put it aside.

 

AI transformation is therefore not simply a technology-adoption exercise. It is an organisational-design challenge.

 

The question is not how many jobs AI can replace. It is what kind of organisation AI allows us to build: one in which machines handle more of the work, certainly, but also one in which people use the resulting capacity to develop deeper expertise, exercise better judgement and take on work that requires capabilities machines do not possess.

 

AI should not simply make organisations more automated. It should make them more capable.

 

---

 

Rebecca Hinds is Head of the Work AI Institute at Glean

 

Main image courtesy of iStockPhoto.com and demaerre