In 1982, Seymour Stein was lying in a hospital bed in New York, waiting for heart surgery, when someone handed him a demo tape from an unknown club singer. He listened once and asked to meet her.
Right there. In the hospital.
She showed up, he signed her, and the world met Madonna.
We love that story because it looks like magic. It wasn’t. Stein had spent decades listening to thousands of demos, signing the Ramones and the Talking Heads before anyone else would and, just as important, passing on acts that went nowhere. His gut wasn’t some magical crystal ball, but a library that he built one wrong call at a time.
I know that kind of library. For twenty years in executive search, I was paid to read people, to know in the first five minutes who would thrive and who would flame out. I was good at it, and I couldn’t explain how. So I went back for a doctorate to find the answer.
Here’s what I found, and here’s what keeps me up at night: we are quietly dismantling the way libraries like that get built.
The science didn’t change. The environment did.
For years, Daniel Kahneman and Gary Klein sat on opposite sides of the intuition debate. Kahneman warned that gut feelings are riddled with bias. Klein showed that experts, like firefighters, nurses, and military commanders, make life-saving calls in seconds. Then, in 2009, they wrote a paper together and discovered they agreed on something essential: intuition can be trusted when it’s built in an environment with regular patterns, lots of practice, and clear, timely feedback.
Practice. Feedback. The freedom to be wrong and find out.
Those are the raw materials of judgment, and you can watch them at work. When Mayo Clinic researchers studied more than 31,000 patient-shifts, they found that 77% of nurses’ “I’m worried about this patient” alerts were confirmed as real deterioration. That’s more accurate than most of the early-warning scores hospitals pay good money for. And nurses with more than a year on the job were more accurate than those with less. That’s what a year of feedback buys you.
In my own research, I interviewed 75 CEOs, CHROs, hiring managers, and executive recruiters about how they make decisions about people. What emerged is a cycle. First comes a physical, emotional, or mental reaction, before you can even say why. Then a pattern match against what you’ve seen before. Then a gut check on how much to trust it here. Then the moment of truth: saying it out loud to someone else. And finally, if you’re disciplined, looking back to ask whether you were right.
Every turn of that cycle is a rep. Every honest answer to was I right? is feedback.
Now look at what’s happening to each of those raw materials. (Spoiler: nothing good.)
The long erosion
It’s tempting to blame AI for all of it. Don’t. AI is the accelerant, not the match.
The reps were already thinning. Economists Natalia Emanuel, Emma Harrington, and Amanda Pallais found that junior engineers who sat near their teammates got far more feedback on their work, and most of that advantage disappeared when offices closed. New hires and younger workers lost the most. The learning didn’t move to Zoom. It just . . . stopped.
Automation was already taking the practice. In 2009, Air France Flight 447 went down over the Atlantic after its autopilot disconnected in a storm. French investigators found that the co-pilots had no training in flying the plane by hand at high altitude. Back in 1983, the engineering psychologist Lisanne Bainbridge had a name for this: the irony of automation. Automate the routine, and you leave humans only the hardest moments, while taking away the very practice they’d need to handle them.
A generation learned to fear being wrong in public. Social media turned every misstep into something permanent and searchable. Jonathan Haidt argues in The Anxious Generation that the move to a phone-based childhood in the early 2010s tracked with a sharp rise in teen anxiety. Whatever the full cause, a generation raised to curate a perfect record learned the exact opposite of what intuition needs: never guess where anyone can watch you miss.
And the experts were already going silent. In my research, 73 of the 75 leaders described using intuition in real decisions. Forty-six told me they were uncomfortable admitting it. Twenty-seven described hiding it, dressing up a hunch as data or keeping it to themselves entirely. I call this selective epistemic suppression, which is a fancy (okay, nerdy) way of saying that people know something, and the room teaches them to hide it. When the people with the best judgment won’t say it out loud, it never gets passed down. The philosopher Michael Polanyi put it simply: we know more than we can tell. Most experts now don’t even try.
Then AI showed up
AI didn’t start the erosion. It turned a slow leak into a flood.
The bottom rung is disappearing. Stanford’s Erik Brynjolfsson and his colleagues found that employment for 22- to 25-year-olds in AI-exposed occupations is now 19% below where it would be had it kept pace with their less-exposed peers, while more experienced workers in the same jobs held steady or grew. But the mechanism matters more than the number. AI is great at replacing book knowledge and lousy at replacing the tacit, earned-on-the-job kind. So it’s eating exactly the work where young people used to earn their reps. As Brynjolfsson has noted, senior workers come from junior workers. Cut the bottom of the ladder, and in a decade, there’s no middle.
Confidence is now on loan. In a study of 319 knowledge workers, researchers at Microsoft and Carnegie Mellon found that the more people trusted AI, the less critical thinking they reported doing. The more they trusted themselves, the more they did. John Nosta, author of The Borrowed Mind, has a name for this: borrowed certainty. The answer arrives sounding sure, so you never build the earned confidence that makes you check it.
Wharton’s Steven Shaw and Gideon Nave take it one step further. They call it cognitive surrender: handing your reasoning over to the machine and adopting its answer in place of your own. In their experiments, once people consulted ChatGPT on a reasoning problem, they went with its answer more than 80% of the time, even when the researchers had rigged it to be wrong. And they felt more confident doing it.
Read that again. Wrong, and surer than ever.
It shows up in the classroom, too. In a randomized trial with nearly 1,000 high school students in Turkey, Hamsa Bastani and her colleagues found that students who practiced math with unrestricted access to GPT-4 scored 17% worse on a later exam than students who never had it. They were also overly optimistic about how much they’d learned. They didn’t just know less. They didn’t know they knew less.
Even experts lose it. In Poland, a study of experienced doctors found that their rate of spotting precancerous growths during colonoscopies, when working without AI, fell from 28.4% to 22.4% within three months of AI assistance being introduced. These weren’t novices who never built the skill. These were experts whose skill was quietly wearing away.
Sure, we’ve heard the panic before. Socrates worried that writing would ruin our memories. Teachers worried about calculators. Our parents worried about GPS. But those tools offloaded memory, arithmetic, and directions, the very things we freed ourselves from so we could focus on judgment.
AI offloads judgment itself.
So what if the real risk isn’t that machines will out-think us?
What if it’s that a generation will grow up with every answer and no feel for when the answer is wrong?
That’s the intuition crisis. And it has already begun.
What we need to do right now
Adam Grant writes that people grow through scaffolding: temporary support that is gradually removed. Here’s the problem. Our scaffolding is vanishing, and AI is a scaffold that never comes off. We can’t stop the technology, and honestly, I wouldn’t want to. But we can rebuild the conditions intuition needs: reps, feedback, and permission to be wrong.
One rule applies to everyone: think first, then prompt. In MIT’s study of AI-assisted essay writing, people who wrote on their own first and then brought in AI did well. Those who started with AI built up what the researchers called cognitive debt. Beyond that, each of us has a different lever to pull.
Leaders: rebuild the apprenticeship
- Design entry-level work for judgment, not output. If AI now drafts the first version, give junior people the job of finding what’s wrong with it, and pay attention to how they do it. The rep isn’t the draft anymore. The rep is the critique.
- Put the error signal back. AI answers don’t come with feedback, so you have to add it. Keep a decision journal for big calls and revisit it. Run a premortem, Gary Klein’s technique for imagining a project has already failed, so people voice the worries they’d otherwise swallow. And in every debrief, ask: What did you sense and not say?
- Make being wrong survivable. A meta-analysis by Nina Keith and Michael Frese across 24 studies found that training that encourages mistakes beats training designed to avoid them, especially on unfamiliar tasks. If a wrong call on your team means public humiliation, people will stop making calls. They’ll just hand them to the machine.
Teachers: grade the reasoning, not the answer
- Let students struggle first. Across 53 studies, Tanmay Sinha and Manu Kapur found that wrestling with a problem before instruction beats getting instruction first. The struggle is where the pattern gets built.
- Choose AI that coaches rather than answers. In Bastani’s trial, a version of the tool built to give hints instead of solutions largely avoided the learning loss. The technology isn’t the problem. The design is.
- Make the thinking visible. When the answer is free, the reasoning is the only thing left worth grading. Ask how a student got there, what they tried first, and where they changed their mind.
Mentors: say the quiet part out loud
- Narrate your hunches. Your instincts are the most valuable thing you can pass on, and you’re probably hiding them. Try: Here’s what I’m sensing, here’s what it calls to mind, and here’s what would change my mind. That’s how a gut gets taught.
- Ask for their read before you give yours. Once you speak, they’ll defer. Hear their instinct first, then compare notes. That comparison is the feedback loop.
- Stay close. The proximity research is clear: feedback happens in passing, in the hallway, in the moment. If your mentee is remote, schedule the passing moments on purpose.
Parents: let them be wrong at home
- Resist the rescue. Jessica Lahey’s The Gift of Failure makes the case that kids whose parents rescue them from every mistake miss the lessons those mistakes teach. Home is the cheapest place there will ever be to get something wrong.
- Match the struggle to the age. The same research on productive failure found that young children, roughly second through fifth grade, did better with instruction first. Little kids need some knowledge before struggle helps. Build the base, then let them stretch.
- Bring “think first, then prompt” to the kitchen table. Before anyone asks the chatbot, everyone guesses. Then check. Then argue about who was closest, and why. (Loser does the dishes.)
Judgment is the last human advantage
Seymour Stein’s hospital-bed call took about three minutes. It took thirty years to build.
The next generation can still build guts like his. But only if someone gives them the reps, tells them the truth about how they did, and lets them be wrong often enough to learn the difference.
Here’s one thing I know to be true: we aren’t running out of answers. We’re running out of people who know when the answer is wrong.
Laura Gassner Otting is a Wall Street Journal bestselling author of Wonderhell, Limitless, and Mission Driven, and a Good Morning America contributor. A former executive search CEO, she spent twenty years placing leaders from the corner office to the Oval Office. Her doctoral research studies how leaders use, hide, and sharpen intuition in high-stakes decisions.