Fifteen Years of Hiring Instincts, Undone in a Year

The instinct that a good hiring manager spent fifteen years building, the gut read that says this one, over the others, was undone in about a year, and most of them do not yet know it. The instinct still fires. It fires as fast and as surely as it always did. That is the trouble. It is now, quietly, pointing the wrong way, because the world it was trained on stopped existing while the reflex kept running.

The hiring managers I work with spent fifteen years learning to feel a strong candidate before they could explain why: the confident answer that had thinking behind it, the portfolio with a mind visible in it, the writing sample where the sentences held their weight. That feel was not mysticism. It was a thousand hiring decisions compressed into a reflex, calibrated on a world where those artifacts cost something to produce. The reflex worked because the collateral was real.

Then the collateral disappeared, roughly between one hiring cycle and the next, and nobody sent the instinct a memo.

The short version:

• The seasoned hiring gut was calibrated on a world where the artifact tracked the capability: a sharp memo meant a sharp mind, because the only way to make one was to have the other. That correlation held for a career. It broke in about a year.
• The instinct now misfires in a specific direction. It rewards fluency, confidence, and polish, once proxies for capability. AI made all three cheap, so the gut now selects for the most surrendered candidate as surely as it once selected for the best one.
• This is not the gut failing. It is the gut working perfectly on data that no longer describes the world. A well-calibrated instrument reading a changed panel gives you a confident, precise, wrong number.
• What replaces the read is a deliberate test the reflex cannot shortcut: a live moment where the machine is confidently wrong, and you watch what the candidate does. Direction, not fluency. We built the AI Operator Profile to measure it.

---

What the instinct actually was

Nobody hires by instinct in the way the word suggests. The gut read is compressed experience, a pattern-matcher trained on outcomes, running below the level where you can narrate it. A manager feels the yes and moves a candidate forward. Ask why and they say "something about how she thought about it." They are right, and they cannot show their working, because the working dissolved into the reflex years ago.

Here is what that reflex was actually keyed to, though the manager never had to know it. It was keyed to cost. You could not sound that fluent about a hard problem unless you had done the thinking. You could not produce those portfolio screens without the mind that solved the problem behind them. The instinct was never reading capability directly. It was reading the artifact, and trusting that artifact and capability moved together, because for a hundred years they did. Faking the signal cost about as much as having the real thing. That was the whole trick, and I walked through it signal by signal in every hiring signal AI can now fake. The instinct was a bet on a correlation. A very good bet. It paid out for fifteen years.

The year the correlation broke

Then a candidate opened a browser tab, and the correlation the whole instinct rested on came apart.

The fluent answer no longer requires the thinking. The clean portfolio no longer requires the mind. The sharp writing sample no longer requires a person who can structure an argument. It requires a person who can prompt one who cannot. AI did not make candidates more capable. It made the artifacts almost free while leaving the capability underneath exactly as rare as it always was. And the instinct, which only ever read the artifact, kept firing yes at the fluent and the polished, with no way of knowing the ground beneath the read had been pulled out.

This is the cruel part. The instinct did not degrade. It works exactly as well as it did five years ago. The problem is that "works" means "maps the input to the output it was trained on," and the input changed meaning underneath it. A compass does not break when you carry it past a magnet. It points as confidently as ever. It just no longer points north. The seasoned hiring gut is that compass, AI is the magnet, and the read still feels dead north.

I traced this same collapse at the level of the whole system in why hiring broke, and it wasn't AI. AI did not invent the problem. It exposed that hiring had always measured proxies, and it destroyed the one property, expensive-to-fake, that made the proxies worth trusting.

Which way the instinct now misfires

A broken instrument that reads randomly is not dangerous. You learn to distrust it. The dangerous instrument reads confidently and consistently wrong in a single direction, because you keep trusting it and it keeps lying the same way.

The hiring gut misfires in exactly that direction. It selects, reliably, for the candidate who leaned hardest on the machine, because that candidate produces the most fluent answer, the most polished sample, the most confident delivery, which are precisely the surface features the instinct was trained to reward. The reflex is not confused. It is doing its job with total fidelity. Its job now hands you the wrong person with the same certainty it once handed you the right one.

There is a measured human tendency underneath this, and it turns the misfire from bad luck into something closer to a law. In a Wharton study titled Thinking, Fast, Slow, and Artificial, researchers Steven Shaw and Gideon Nave ran three experiments with 1,372 people on reasoning tasks. Accuracy rose from 46% with no AI to 71% when the AI was right. Then, when the AI was wrong, accuracy fell to 31.5%, below the 46% people managed with no help at all, and they followed the wrong answer around 80% of the time, their confidence rising even as they got it wrong. Shaw and Nave call it cognitive surrender: the mind handing the thinking to the machine and borrowing its certainty without checking whether it was right.

Read that next to the hiring gut and the trap snaps shut. The candidate who surrendered most completely feels most confident, because they inherited the model's certainty and never spent a doubt on it. The candidate who actually thought, who argued with the output and caught it, presents with less polish, because thinking leaves seams and surrender does not. So the instinct, reaching for confidence the way it always has, now reaches past the person who reasoned and lands on the person who abdicated. The read is not just unreliable. It is inverted.

The old instinct, and where it now points wrong

The old instinct	The signal it read	Why it worked	Why it now misfires
"She's sharp"	A confident, fluent answer to a hard question	Fluency on a hard problem required doing the thinking first	The model supplies the fluency; confidence now tracks surrender, not reasoning
"He can clearly do the work"	A clean, finished portfolio or sample	The artifact could not exist without the mind that made it	The artifact is near-free to produce; it proves the tool works, not the person
"Good judgement, listen to how she reasons"	A crisp, well-structured verbal answer	Structure under light pressure signalled a structured mind	The structure can be pre-cooked from a model's output; polish no longer implies the reasoning happened

The quiet cost, one hire at a time

One misfired read is a bad hire and a hard quarter. A hiring process still run on the old instinct is worse: an organisation that systematically selects for surrender and calls it talent, and cannot see it happening because every individual decision felt right.

Watch what a team becomes twelve months into trusting the broken compass. The people who got hired are fluent, confident, fast. They interview beautifully, because interviewing beautifully is exactly the artifact AI now makes cheap. Output looks healthy. And underneath, the share of work that is machine-shaped and unchecked climbs, because the process filtered for people who trust the model and filtered out the ones who argue with it. That is the collapse of talent signals arriving inside a single company, one confident hire at a time. The metrics say you are getting stronger while your capacity to catch a wrong answer quietly rots, discovered only when a confidently wrong decision reaches a customer or a P&L.

The macro data rhymes with the micro. PwC's AI Jobs Barometer finds AI-related skills now sitting on roughly 2.5% of job postings and climbing. Everyone is screening for fluency. Almost nobody has changed what the fluency now means.

What replaces the read

You do not fix a broken instinct by trying harder to trust it, and you do not fix it by throwing it away. Fifteen years of pattern-matching is not something you can switch off, and the parts of it that read character, drive, how someone treats the room are still keyed to a world that hasn't changed. The judgement that broke is narrow and specific: the read that inferred capability from a producible artifact. That one read has to come off the gut and go onto a deliberate test.

The test has one shape, because only one shape survives. It happens live, in front of you, on a problem the candidate cannot pre-cook, at the moment the machine is confidently wrong. Hand them a real task with an AI output already attached, polished, plausible, and quietly incorrect, and say nothing about the error. Then watch. The surrendered candidate, the one your old instinct adores, accepts the gift and builds on it. The other kind gets an itch, checks the load-bearing claim, finds the crack, and rebuilds the part that was rotten. Same tool, same task, same clock. Opposite person. That is the fluency-versus-direction split, and it is the whole hire, the case I made in stop hiring AI users, start hiring AI Operators.

This behaviour cannot be faked, for one clean reason: faking it would require the exact judgement you are testing for. To pass, the candidate has to know the AI is wrong, know where, and know how to fix it, which is precisely the capability that makes them worth hiring. Passing the test is the thing the test measures. Your gut cannot read this from a résumé, because it does not live in an artifact. It lives in a live minute you have to deliberately build.

So the elegy is real, and it is also smaller than it feels. A trusted instinct went wrong, genuinely wrong, not dulled but inverted, and that loss deserves to be named rather than papered over with a new tool and a brave face. But what it took was one correlation, not the whole craft. The read that inferred a mind from a portfolio is gone and is not coming back. The judgement of the person underneath the portfolio is exactly what the next fifteen years will be about. You just have to stop letting the machine answer the question for you.

Ivanooo built the AI Operator Profile for the read the instinct can no longer make on its own: not what a candidate can make AI produce, but whether they can direct it, and catch it, when it is confidently wrong.

---

Frequently asked questions

Why is my hiring instinct suddenly unreliable when it worked for years? Because the instinct was never reading capability directly. It was reading artifacts, the fluent answer, the clean portfolio, the sharp sample, and trusting they moved together with capability, which for a career they did. AI made those artifacts cheap to produce without the underlying capability, so the correlation your gut relied on broke. The reflex still fires accurately, just at signals that no longer mean what they meant.

Does this mean experienced recruiters are now worse at hiring than beginners? No. One specific part of the seasoned read, inferring capability from a producible artifact, has inverted, and experience makes that part fire faster and more confidently, which is the risk. The rest of a recruiter's judgement, reading character, drive, how a person handles the room, still works. The fix is to move the artifact-read off instinct and onto a deliberate live test, not to distrust the whole craft.

Why would my gut now favour a weaker candidate over a stronger one? Because it reaches for fluency and confidence, and AI made those the cheapest to fake. The candidate who leaned hardest on the machine presents as the most polished and the most certain. Wharton's Shaw and Nave found people grew more confident even as AI led them to wrong answers around 80% of the time. Thinking leaves seams; surrender looks smooth. So the instinct lands on the surrendered candidate and reads their borrowed certainty as strength.

What actually changed in "about a year"? The cost of producing a convincing artifact collapsed. A résumé tuned to your keywords, a structured writing sample, a finished-looking take-home, all of these used to require the capability they signalled. Between roughly one hiring cycle and the next, general-purpose AI made them producible by someone without that capability, severing the link between the artifact and the person.

How do I test for the capability my instinct can no longer see? Build a live moment the candidate cannot pre-cook. Give them a real task with an AI answer already attached that is plausible but quietly wrong, say nothing about the error, and watch. The ones who catch it, check the load-bearing claim, and rebuild are showing direction, the judgement that survives the tool. The ones who build on the wrong answer are showing fluency alone. Faking a pass requires the exact judgement you are testing for, which is why the signal holds.

Can't I just add AI-detection to my screening and keep trusting my read? No, and it is the wrong fight. AI-assisted work is now the median candidate's default, so detection would flag almost everyone and still tell you nothing about who can direct the model versus who surrenders to it. The problem is not that some artifacts are AI-shaped. It is that all of them now are. You change what you measure, behaviour under a live wrong answer, rather than trying to restore trust in an artifact that no longer carries the signal.