You use AI to keep up.
Is it quietly dulling your edge?

You finished the deck in half the time. The draft was clean, the analysis held up, nobody in the room could tell how you got there. And still, somewhere on the drive home, a small question surfaced that you did not quite want to look at: could you have done that on your own anymore? Not the typing. The thinking. The part you used to be known for.

If that question has started visiting you, you are not being dramatic, and you are not alone. It has a name in the research, and two studies published in 2025 put a worrying shape around the feeling.

Here is the short answer, before the detail.

Leaning on AI does not erase your skill. It quietly stops you using it, and your confidence fades in step with the practice you stop getting.

Can relying on AI actually make you worse at your job?

In a specific and measurable way, it can. The mechanism is not that AI reaches into your head and removes anything. It is that every time you let the tool do the reasoning, you skip a rep you would otherwise have done yourself. Do that a hundred times a week and the muscle you built over a career gets less use than it has had in years. Skill that stops being practised does not vanish overnight, but it does go quiet, and quiet skill is very hard to trust when you need it under pressure.

Researchers call the mechanism cognitive offloading: handing a mental task to an external system instead of holding it yourself. It is not new. You stopped memorising phone numbers the day your phone started doing it for you, and you lost nothing important. What is new is the kind of task now being offloaded. Not remembering a number. The actual work of forming a view, weighing evidence, and deciding what you think.

What did the 2025 research on AI and skill actually find?

A study by Michael Gerlich, published in January 2025 in the journal Societies, surveyed 666 people across a range of ages and backgrounds. It found a significant negative correlation between frequent AI tool use and critical thinking ability, and the link ran through cognitive offloading. The more people trusted AI and let it carry the thinking, the less they engaged in critical thinking of their own. Younger participants, who leaned on the tools most, showed the steepest drop.

A second study, run by Microsoft Research together with Carnegie Mellon University and published in early 2025, went straight at working professionals. It surveyed 319 knowledge workers about 936 real tasks they had done with AI. The finding that should stop you: the more confidence a worker had in the AI, the less critical thinking they applied to what it produced. The professionals who kept thinking hardest were the ones who still trusted their own judgment more than the tool's.

The same study noticed the work itself changing shape. Across the tasks people described, AI shifted their effort away from doing and toward overseeing: less problem-solving, more checking the machine's answer. That is a quieter job than it sounds. Supervising output you no longer fully generate is a strange, weightless kind of work, and it is easy to feel less like the author of your day and more like its editor.

Why this lands harder on experienced professionals

If you are early in a career, offloading a task you never mastered is a smaller loss. If you are fifteen or twenty years in, the calculation is different, because the judgment you are quietly setting down is the thing you spent those years building. It is not just a skill. It is the evidence, to yourself, that you are good at what you do.

That is why this so often shows up not as a productivity worry but as a self-worth one. A 2025 technostress study published in Frontiers in Psychology found that the specific fear of AI surpassing skills built over years of training and experience was linked to measurably higher anxiety and depressive symptoms, with techno-insecurity correlating with anxiety (r = 0.233) and with depression (r = 0.244). The threat the body registers is not only losing a task. It is watching the ground your identity stands on start to soften.

This is close cousin to the dread that clinicians have started calling AI Replacement Dysfunction, the fear of professional obsolescence, and to the older, more familiar feeling of being a fraud at work. The difference is that this one comes with a grain of real evidence, which is exactly what makes it stick. When the doubt has a factual footing, telling yourself you are imagining it does not work.

The confidence trap hiding inside all this

Read the two studies together and a loop appears. The more you trust the tool, the less you practise your own thinking. The less you practise, the less you trust your own thinking. The less you trust it, the more you reach for the tool. Each turn of that loop feels efficient in the moment and costs you a little more of the thing you cannot outsource: the felt sense that you know what you are doing.

The Microsoft and Carnegie Mellon result is really a description of where that loop can be broken. The workers who stayed sharp were not the ones who avoided AI. They were the ones who went in with enough confidence in their own view to argue with the output. Self-trust was the variable that protected the thinking. Which means the goal is not to use AI less out of fear. It is to keep enough of your own judgment in play that you are still a party to the decision, not just its rubber stamp.

AI did not take your job. It quietly took a hundred small decisions you used to make yourself, and your confidence left with them.

What actually helps if AI is eroding your confidence at work?

The first move is honest visibility: which parts of your work have you actually handed over, and does it matter that you did? Some tasks are fine to offload forever. Others are load-bearing to your edge, and those are the ones to protect. It is hard to tell the two apart while you are moving fast, which is what the free AI Exposure Calculator is built to surface: where your work overlaps with what AI now does well, and which of your skills is worth deliberately keeping warm. The Cognitive Bandwidth Index looks at the same picture from the inside, at how much genuine thinking room you have left once the constant tool-switching and oversight are accounted for.

The second move is a change of posture, not of tools. Treat AI as a sparring partner rather than an oracle. Form your own rough view first, even a bad one, then bring the machine in to pressure-test it. Keep a few tasks deliberately hands-on, not because it is efficient, but because it is the practice that keeps your judgment yours. The point is not to work harder. It is to stay in the room while you work.

And the third, quieter piece is the one the calculators cannot reach. Underneath the productivity question is usually an identity question: if the tool can do the thing I was known for, then who am I now at work, and what is still mine to offer. That is a real question, not a neurotic one, and it does not resolve by reading another article about AI. It resolves in conversation, out loud, with someone who is not inside your career and not trying to sell you a certainty. Being able to say the fear plainly, and hear yourself think past it, is often where the ground starts to feel solid again.

Common questions about AI, skill, and confidence at work