Essays · Leadership · AI and work · Psychology

Confidence you build versus confidence you borrow, for leaders

AI can make people feel capable fast. Here is how leaders tell built confidence from borrowed confidence, and why it belongs on the balance sheet.

The most patient voice in many people’s day is now a machine. It never sighs. It never checks the clock. It answers the eleventh question as kindly as the first.

That patience is a real gift, and I do not want to wave it away. I have seen learners light up because a piece of software finally explained something without impatience. But it raises a question every leader needs to answer right now: the confidence your people feel, is it backed by anything?

I come at this from two chairs. As a school psychologist, I sit across small tables from students who have spent years quietly carrying a belief that they are “not a math person” or “not smart.” As a builder of AI-powered education platforms, and a former senior manager at Oracle, I make decisions about what software should say to someone who is stuck. From both chairs I have reached one conclusion. The biggest question about AI is not whether it gives the right answers. It is what it does to people’s belief in themselves, and whether that belief is connected to anything real.

What confidence really is

Albert Bandura, the Stanford psychologist who spent decades studying it, called it self-efficacy: your belief that you can do a specific thing well enough to get the result you are after.

It is not self-esteem, and it is not a confident personality. You can feel great about yourself and still believe you cannot run a budget. You can be quiet in meetings and completely sure of yourself in a spreadsheet. Bandura showed that this specific belief shapes what people try, how hard they push, how long they stick with it, and how fast they bounce back.

He also found it comes from four places. In When Machines Believe in You, I organize them as four accounts in what I call the Confidence Ledger:

  • Mastery: what you have actually done.
  • Models: whom you have watched succeed.
  • Messages: what you are told about your ability.
  • Mood: how your body and emotions read the challenge.

The ledger has one rule. Every deposit of encouragement has to be backed by evidence of real competence, or it is counterfeit currency.

The account machines cannot fill

The accounts are not equal. Machines are very good at Messages (“Great work”) and at Mood, lowering the threat that keeps people from trying. The strongest account, Mastery, is the one a machine cannot deliver at all. It has to be earned.

Here is the twist most people miss. Bandura warned that easy wins build fragile confidence. People who only experience quick, easy successes come to expect quick results, and the first real setback knocks them flat. The mastery that holds up comes from succeeding at something after it resisted you.

Every athlete knows this. Game-day confidence does not come from the pep talk in the tunnel. It comes from the reps you sweated through in practice when nobody was watching.

So when an AI drafts the report while someone watches, the work got done, and that has value. But nothing resisted them. The resistance got handled somewhere else. Easy help does not count toward the Mastery account. That is the difference between confidence you build and confidence you borrow.

Borrowed competence stays home

There is a simple way to tell the difference, and it works on yourself, your kids, or your team. I call it the Tool-Off Test.

  1. Can you explain it? Close the tool and explain the idea out loud in your own words. Where you get stuck is the part you borrowed.
  2. Can you do it? Take a fresh problem and do it from scratch. It does not have to be fast. It has to be yours.
  3. Can you use it somewhere new? Change something: a different client, a harder version. Real competence travels. Borrowed competence stays home.

Confidence is a balance-sheet item

Here is the idea I want every leader to take into Monday. In the AI era, your people’s confidence is a balance-sheet item. Built confidence is an asset that compounds. Counterfeit confidence is a liability you have not booked yet, and it tends to come due all at once.

Most organizations measure AI by adoption, speed, satisfaction, and self-reported confidence. Every one of those numbers can rise while actual capability falls. Researchers at BetterUp Labs and the Stanford Social Media Lab named one result of that gap in 2025: workslop, AI-generated work that looks polished but does not do the job. In their survey of about a thousand American desk workers, four in ten said they had received some in the past month, and each instance took nearly two hours to deal with.

And as of late summer 2026, McKinsey’s annual survey found that nearly nine in ten companies were using AI regularly, but only just over a third could point to measurable impact on profit. There are many reasons for that gap. I would argue one of them is that companies are buying output and assuming they are getting capability.

What leaders can do

Most AI decisions can be delegated. A few cannot.

Measure one thing that only rises if people are getting better. Adoption tells you people are using the tool. It does not tell you they are improving. Add at least one measure of independent capability.

Decide how new people learn. When AI handles the routine work junior people used to do, it also removes the way they used to learn it. Choose, on purpose, which tasks new hires still do by hand at first, and who they shadow.

Decide what managers are for. Let the tools answer “How am I doing?” Make managers responsible for “Where am I going?” Then give them the time to have that conversation.

Interview the reasoning, not the result. When a work sample looks strong, spend the interview inside it. Why this structure? What is the weakest part? People who did the thinking light up. People who did not get vague.

Say it out loud. Something like: we expect you to use AI, and we also expect that six months from now you will be able to explain everything you hand in. If those two ever conflict, the second one wins.

Belief before evidence

The ledger can sound colder than I mean it. Sometimes belief has to come before evidence. Someone who has never succeeded at something has no mastery to point to yet, and somebody has to extend credit.

Think of the coach who tells a gangly ninth grader he is going to be a point guard, a full year before the kid can dribble with his left hand. That is a loan of belief, and great leaders make those loans all the time. The point is that a loan has to be paid back in real reps. Believe in people early, and then build the practice that turns your belief into their evidence.

Machines can supply the patience. People still have to supply the belief, the standards, the witness, and the evidence. Someone has to mean it.

The next play

  • Add one measure of independent capability to the way you track AI use on your team.
  • Run the Tool-Off Test on yourself this week, on one piece of work you produced with AI.
  • In your next one-on-one, skip “How did it go?” and ask “Where are you headed?”
  • Give one person a receipt instead of a cheer: name the specific step they did themselves.
  • Extend one loan of belief to someone who has not earned it yet, and plan the first rep with them.