The more your team leans on AI, the worse they get at the job. That's now measured, not theoretical.

Your best people are faster than they have ever been, and most of the time the AI-assisted work is genuinely good. But you may have started to notice something you cannot quite put your finger on: when the tool is unavailable, or when a genuinely unusual problem lands on the desk, there is a hesitation that was not there before. The technology made them quicker. You are beginning to wonder whether it is also, underneath, making them a little less capable. That worry now has evidence behind it.

The clearest example comes from medicine. A 2025 study in The Lancet Gastroenterology and Hepatology, across four centres, followed what happened to endoscopists once AI polyp-detection became routine. Their ability to detect adenomas without the AI fell from 28.4% before exposure to 22.4% afterwards, a decline of around 20%. It is the first real-world evidence that habitual reliance on AI can erode the underlying human skill, and here it did so in a setting where the skill matters for whether cancers get caught.

Isn't this just a medical curiosity?

No, because the same pattern shows up in knowledge work. A study by Microsoft and Carnegie Mellon of 319 knowledge workers found that the more confidence people had in generative AI, the less critical thinking they applied to its output. AI reduces the perceived effort of thinking, and the effort you stop spending is precisely the skill you stop keeping. The mechanism is not exotic. It is the oldest rule of human capability: use it or lose it. AI moves people from doing the work to supervising it, and supervision without practice quietly wastes the muscle you were supervising with.

This is the shadow side of the shift from hands-on work to overseeing AI, the same shift explored in the move from prompting to agentic systems. Handing the doing to the machine is the point. The unintended cost is that the humans slowly lose the ability to do it themselves.

Why does deskilling matter if the AI is usually right?

Because "usually" is exactly the trap. You still need the human skill for the moments the AI cannot cover: when it is wrong, when it is unavailable, when the problem is genuinely novel, and for the judgement to tell which of those you are in. A team that has quietly deskilled cannot catch the AI's mistakes or take the controls when it fails, which is precisely when you need them sharp. And there is a slower danger in the pipeline: people who lean on AI from their first day may never build the deep competence that lets an experienced person supervise AI well. You can deskill the next generation before it has ever been skilled.

How do leaders keep the skill alive?

By treating capability as something you protect on purpose, rather than assume survives.

1. Notice what you are outsourcing wholesale. The skills handed entirely to AI, with no human practice left, are the ones quietly at risk. Name them.
2. Keep people practising the judgement, not just approving output. Rubber-stamping AI is not practice. Reserve real decisions for humans to actually make.
3. Build "AI-off" reps. Like pilots who still hand-fly, schedule periodic unaided practice so the core skill stays live and ready.
4. Reward the challenge, not just the speed. Make questioning and catching the AI part of the job people are valued for, not friction to be smoothed away.
5. Protect the training pipeline. Make sure juniors build genuine skill before they are allowed to lean on AI for it, or you lose your future supervisors.

What does this change for me as a leader?

It makes human capability something you steward rather than take for granted. AI's speed is real and worth capturing; the risk is that you bank the speed while quietly spending down the skill that made your people valuable in the first place. The organisations that come out ahead will keep both, using AI to augment the work while deliberately keeping their people sharp underneath it, so that when the moment comes that only a skilled human will do, there is still one in the room.

This is the human core of the end of business as usual: as the tools take over more of the doing, the deliberate cultivation of human skill becomes a leadership responsibility rather than a given. Keep the muscle in use, and AI makes your people more capable rather than less. It is part of what it means to become genuinely AI-native: fluent with the tools, and still formidable without them.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.