When AI Does Your Thinking, You Lose the Reps That Build Conviction

In CNBC Make It, IBM exec: How managers can help prevent employee skill erosion from AI looks at a worry spreading quietly through the workforce. Aparna Nair, IBM's chief talent, leadership and culture officer, argues that managers are best placed to help because they know both the work and the individuals doing it. The piece also names the complication: many companies are cutting the very management layers that would do this work.

The data behind it comes from IBM's 2026 CHRO study, which surveyed 1,500 CHROs and 8,800 employees. Sixty percent of employees worry AI is eroding their skills, and critical thinking is the skill they most often see declining. Among those who are worried, three in four say the erosion has already started.

Borrowed Answers, Borrowed Confidence

The coverage treats this as a skills problem, and it is one. The deeper loss is harder to measure. Critical thinking is more than a line on a résumé. It is how people come to trust their own judgment, and it is how they find out what they actually believe.

Researchers at the Wharton School have a name for the pattern: cognitive surrender, accepting an AI's answer in place of your own reasoning. Each time it happens, a person skips a rep. One skipped rep costs nothing. Thousands of them change who you are at work. You become someone who can produce an answer but can't defend it, adjust it or tell when it's wrong.

I describe conviction as the earned capacity to act on what you believe before the outcome is certain. The word that matters is earned. Conviction is built through repetition: forming a view, testing it, getting it wrong and forming a better one. Confidence that comes from an AI's output is borrowed. It looks the same in a meeting, right up until someone asks the second question. I made a similar case in AI Can Fake the Catch. It Can't Fake What You Earned. The tools have only become more capable since then, which makes the point more urgent.

The Manager Is the Missing Piece

Nair's answer is the right one, and it's worth taking further. No single training program will solve this, because erosion looks different in every person. Common training and clear guidelines for AI use are valuable. They give everyone a shared starting point and a consistent standard. But the analyst who has stopped questioning the model, the new hire who never built the skill in the first place and the veteran whose judgment is quietly going unused each need something different. Only someone who knows them can tell which is which. A manager who has watched someone work for a year can see the difference between a person who is learning faster with AI and one who is quietly leaning on it.

IBM's study also points to the environment around people. Where HR shares responsibility for deciding which decisions stay human-led, 76% of employees feel safe questioning or overriding AI recommendations, compared with 43% where HR is only advisory. People exercise judgment when they believe it is wanted. If a culture rewards speed and treats disagreement with the machine as friction, people will stop disagreeing. Their thinking then weakens because the room taught them it wasn't needed, not because they stopped caring.

That is why flattening management to save cost carries a hidden price. The layers being removed often include the people whose job was to know each person well enough to notice when they are slipping. I explored this in AI Can Learn Your Expertise. It Can't Learn Who You Are. What AI can't replicate is the individual, and organizations that remove the people who know their individuals will eventually feel the cost.

Three Ways to Protect the Reps

For managers: in your next one-on-one, ask someone to walk you through how they reached a recent conclusion before you look at the output. You will learn quickly whether the thinking is theirs, and they will learn that you care about how they think.

For teams: make overriding AI a normal part of the work. Ask in your weekly meeting where the team disagreed with the tool and why. Recognize the good calls out loud, so people see that judgment is valued.

For individuals: before you prompt, write down your own first answer. Then compare it with what the tool gives you. That one habit keeps you in the reps and shows you where your judgment is strong and where it needs work. The goal is to keep using AI while remaining the author of your own thinking.

The professionals who thrive in the next decade won't be the ones with the most answers. They will be the ones who still know how they got there, and who trust themselves enough to say so.

Answers you were handed build borrowed confidence. Conviction is earned in the reps you refuse to skip.

Want to explore these ideas further? Learn more about my work on leadership, identity and conviction at www.theglennllopis.com.