When AI Makes You Worse at Work

Part five, and the last, of a series. This one follows the habits to where most of us spend our days: work.

So far this series has made an argument, shown the evidence, given adults a playbook, and turned to kids. This final piece is about your career, and it carries a warning that is easy to miss because the early days of a new tool feel so good: the same AI that makes you faster this quarter can make you worse at your job over time, if you let it do the part of the work that was keeping you sharp.

Can AI actually make you worse at your job?

It can, and we now have a clear case of it in a field where the stakes are about as high as they get. In a 2025 study in The Lancet Gastroenterology & Hepatology, researchers tracked endoscopists at four centres before and after those doctors began using AI to help spot polyps during colonoscopies. When the doctors later worked without the AI, their detection rate had fallen from 28.4% to 22.4%. The skill they had handed to the machine had measurably faded in a matter of months. It was an observational study rather than a controlled trial, so treat it as a strong signal rather than a final verdict. But it is the first hard documentation of trained professionals being de-skilled by their own AI, at the exact task they are paid to be good at.

Why does relying on AI erode skill?

This is not new, and it is not really about AI. Back in 1983, the psychologist Lisanne Bainbridge described what she called the ironies of automation. The more a system does for you, the less you practice doing it yourself, so your skill fades. And then you are handed the hardest version of the job, the moment the automation cannot cope, with the least practice to meet it. You are expected to supervise a system you have slowly lost the ability to do without.

Aviation learned this the hard way. Autopilot handles routine flight so well that pilots can lose the feel for flying by hand, and the case most often cited is Air France 447 in 2009, where the autopilot disengaged in a storm and the crew struggled to recover an aircraft they were no longer used to flying manually. Bainbridge’s own conclusion is the one worth carrying into the AI era: the most automated systems may need the most operator training, not the least. The tool does not remove the need for skill. It hides it until the day it is needed most.

AI does not make you redundant. Letting your judgment rust does.

What does leaning on AI do to your thinking at work?

It shifts where your effort goes, and not always for the better. A 2025 survey of 319 knowledge workers by Microsoft and Carnegie Mellon found that the more confidence people had in the AI, the less critical thinking they reported doing. The work shifted from producing the thing to checking the machine’s version of it, which is fine when you actually check, and a problem when deadline pressure means you skim and ship. Because it is self-reported, treat it as a signal, not proof. But it lines up with everything else here, and with the cognitive offloading idea that runs through this series: the mental work you stop doing is the mental work you slowly lose.

Who is most at risk?

Early-career workers, by a wide margin, and this is the part that should worry every industry. Expertise is built by doing the unglamorous entry-level work: the research memo, the first draft, the basic analysis, the code nobody wants to write. That grind is how a junior turns into a senior who can tell, at a glance, when something is off. AI now does exactly that entry-level work, faster and for free. The short-term win is obvious. The long-term cost is a generation that never built the judgment the senior people take for granted, because the machine did the reps that used to build it. A field that lets AI do all the apprentice work has to ask where its next set of experts is going to come from.

How do you stay sharp as a professional?

The same way pilots are supposed to: keep flying by hand often enough that you still can. The habits piece has the full version, but at work it comes down to where you point the tool. Do the core judgment yourself and let AI handle the edges, not the reverse. Here is the test, in the language of the job.

Keeping you sharp	De-skilling you
You do the core thinking; AI handles the edges	AI does the core thinking; you rubber-stamp it
You check the AI against your own reasoning	You accept the output because the deadline is close
You could still do the task without it, just slower	You could not do the task without it at all
You work unaided often enough to keep the skill	You have not done the work without AI in months

The single most useful habit is the manual-flying rep: every so often, do a core task the old way, without the AI, just to confirm you still can. If you cannot, that is not a scheduling problem. That is the warning light.

What should managers do about it?

Managers feel this last, because the costs are invisible until they are not. Output looks great right up until the day you need senior judgment that nobody on the team developed. So treat de-skilling as a real risk to manage, not a worry to wave off.

So is AI bad for your career?

No. Used well, it is one of the largest advantages a worker has ever had. The professionals who win with it are not the ones who hand it the most work. They are the ones who stay good enough at the work to aim it, catch its mistakes, and take over when it is wrong. AI used on top of real skill makes you formidable. AI used in place of real skill makes you replaceable. The whole choice is which one you are building.

That is where this series lands, across reading, attention, children, and work. The technology is remarkable and worth using. The thing worth protecting is the human underneath it: the reading, the focus, the judgment that lets you direct the tool instead of depending on it. Use AI to do more of your best thinking. Never to do less of it.

Common questions

Does this mean I should not use AI at work?

Not at all. Use it. The risk is narrow and specific: do not let it do the core judgment that is actually your job. Point it at the routine and the rough drafts, keep the decisions and the checking for yourself.

Isn’t some de-skilling fine?

Yes. Almost nobody can still do long division by hand or read a paper map, and we are fine. The line is whether the skill is central to your work. Losing a skill you no longer need is progress. Losing the one you are paid for is a problem.

I’m junior. How do I build skill if AI does the entry-level work?

Do the work yourself first, then compare it against what the AI produces and study the gap. Ask the AI to explain its reasoning rather than just hand you the answer. Treat it as a demanding tutor, the way the habits piece describes, and you get the speed without skipping the learning.

How do I know if I’m being de-skilled?

Try doing a core task without the AI. If you find you cannot, or that it now feels much harder than it used to, you have your answer. That discomfort is the signal to start taking the manual-flying reps seriously.

Sources

• STAT / The Lancet Gastroenterology & Hepatology: AI de-skilling in colonoscopy (Budzyn et al., 2025)
• Lisanne Bainbridge, “Ironies of Automation” (1983)
• Microsoft & Carnegie Mellon: generative AI and critical thinking (2025)

You may also like

• The Tool That Uses You (part one)
• The Reading Crisis Before AI (part two)
• Use AI Without Losing Your Edge (part three)
• AI and Your Kid’s Reading (part four)
• Glossary: What Is Cognitive Offloading?
• What AI Literacy Means
• The full AI and the Human Mind series
• The daily Beginners in AI newsletter