Selection

IThe Word We Use

The most honest moment in any hiring process is the debrief, and the most honest word in the debrief is bet.

Listen for it. It surfaces after the structured questions are exhausted and the scorecards are filled in, when the panel stops performing rigor and starts saying what it actually thinks. I’d bet on her. He’s a bet, but a good one. I don’t know if she can do the job today, but I’d bet on where she’s going. We reach for the language of wagering because that is what the work has quietly become: not the staffing of a role but the pricing of a future.

This should be stranger than it feels. The official story of hiring is a measurement problem. There is a job, the job requires capabilities, the interview exists to find out whether the capabilities are present. Sort the qualified from the unqualified, match the person to the slot. On that account the right verb is assess. Nobody who has sat on a real panel uses it.

Because nobody believes the job will stay still long enough to be measured against. The role will mutate within eighteen months. The tools will be replaced. The function may be automated, merged, or eliminated. So the question is no longer can this person do the job. It is will this person still be worth betting on after the job is gone — a guess about a person’s whole trajectory through a changing world, not a reading of their present skill.

And in an economy that is pulling apart, a guess about whether someone will stay valuable is a guess about which side of the divide they will spend their life on. The debrief never puts it that way. But that is the wager underneath it: not which seat to fill but which fate to underwrite — and then, by hiring or not hiring, helping the fate come true.

The thought leads somewhere frightening, and I am implicated in it about as completely as a person can be. I work in AI. I sit on the panels. I say the word.¹

IIThe Line

Start with the shape of the thing we are predicting about.

For most of the postwar period the American economy grew like a rising tide — uneven, unfair in a hundred documented ways, but moving in one broad direction. That is no longer the shape. The shape now has a name that began as a description of the pandemic recovery and hardened into a description of the structure itself: the K. One arm rises. The other falls. They start together and pull apart, and the pulling apart is the whole story.²

The numbers are not subtle. The wealthiest one percent of American households now hold roughly a third of the country’s wealth. The bottom half of the country owns something like one percent of the stock market, which means that when the market sets record after record, as it has, the gains land almost entirely on people who were already holding the assets.³ And labor’s share of GDP — the slice of the whole economy paid out to workers as wages, stable for decades — recently fell to about the lowest level on record.⁴ The economy is booming and workers are taking the smallest cut of it ever measured. Both things are true because they are the same thing. That is the K.

If the story stopped there it would only be inequality, which is old news, and which the country has historically tolerated on the theory that the bottom arm can climb to the top one. The American settlement was never equality of outcome but mobility — the promise that the line you were born on was not the line you would die on. As long as that held, your current position was a way-station and a bad starting hand was survivable, because the deck got reshuffled as you played.

The promise has quietly failed. The cleanest measurement comes from a team led by the economist Raj Chetty, who asked a simple question: what fraction of children grow up to out-earn their parents? For children born in 1940, about ninety percent. For children born in the 1980s, about fifty. A coin flip. For sons measured against their fathers, the collapse is starker still — from nearly universal to around forty percent.⁵ The ladder didn’t get harder to climb. It got removed, one rung per decade, and replaced with a coin.

And here is what turns it from sad to structural. Chetty’s team ran the counterfactuals to find out why mobility died, and the answer was not slowing growth. Restore midcentury growth rates and you recover almost none of the lost mobility. What killed the climb was distribution — the fact that the growth, whatever its rate, now pools at the top of the K instead of spreading across it. Roughly seventy percent of the decline traces to where the growth goes, not to how much of it there is.⁶

Now apply that to hiring. There is no longer enough movement in the system for a person’s position to be temporary. The thing that used to make a hiring decision survivable — for the candidate, and frankly for the conscience of the person doing the hiring — was mobility. Didn’t get the offer, got sorted onto the lower arm? You’d climb. The decision was a moment, not a sentence. As mobility dies, the decision stops reversing. Where you get sorted is, increasingly, where you stay. Hiring used to be triage. It is becoming assignment.

IIIWhat We Are Actually Measuring

So we are placing a wager about which arm of a diverging economy a person lands on, and the wager is increasingly permanent. Which raises the question the debrief never asks: when we place that bet, what are we actually reading? What is the signal we believe predicts the upper arm?

We tell ourselves we are reading skill. The most rigorous look anyone has taken at how elite firms hire says otherwise.

The sociologist Lauren Rivera spent years inside the recruiting processes of top investment banks, law firms, and consultancies — the firms that function as on-ramps to the upper arm — interviewing the people who make the calls and watching a hiring committee work.⁷ What she found is that hiring at these firms is not primarily skill-sorting but cultural matching. Evaluators consistently favored candidates who resembled themselves — not in competence, but in leisure pursuits, self-presentation, the texture of their backgrounds, what her subjects called "play." They were looking, more or less openly, for people they would want as friends. And these concerns about similarity routinely outweighed concerns about raw productivity. The thing evaluators screened for most often was not ability but likeness.⁸

Read that against the shape of the K and the machinery becomes visible. The traits that register as "high-potential," "polished," "good fit," "someone we’d want in the room" are largely the traits of people already standing on the upper arm — the easy leisure, the frictionless polish, the fluency that comes from growing up among people who already won. A candidate from the lower arm, the one who believed the official story and poured everything into measurable achievement instead of squash and gap years and the right kind of small talk⁹, reads as narrow. As a grind. As, in the actual word Rivera’s evaluators used about such people, a nerd.¹⁰ We are not, mostly, choosing who can do the work; we are choosing who resembles the people who already did it — and calling the resemblance merit.

This is the move the philosopher Michael Sandel has spent a career trying to make people see. A meritocracy, he argues, is less a system for rewarding ability than for laundering it — taking an inherited distribution of advantage and converting it into a story about desert, so the people on top can believe they earned the arm they were born on, and the people below can be told their position is a verdict on their worth.¹¹ The legal scholar Daniel Markovits puts it more bluntly: the whole apparatus is built, he writes, to "launder an otherwise offensive distribution of advantage."¹² The genius of the system is that it feels like fairness from the inside. The more perfectly meritocratic the selection appears, the more total the winners' conviction that they deserve the spoils — and the more crushing the losers' conviction that they deserve the lack.

Now stack the three findings. Cultural-matching bias, in a high-mobility economy, was unjust but survivable: passed over for resembling the wrong people, you got another hand. In a dead-mobility, K-shaped economy it curdles into something worse — a hereditary sorting mechanism in the costume of merit. We screen for "fit" and "potential," signals soaked in who already made it, to decide who joins the people who already made it, in a system where that decision no longer reverses. And we experience the whole thing as fair.

IVThe Machine Hiring for the World the Machine Is Making

That is the general condition. Mine is more specific, and it contains a recursion I can’t stop seeing.

I work in AI.¹³ The technology I help bring to market is, by the open testimony of the people building it, one of the primary forces widening the same line. That isn’t my inference; it’s the stated position of the field’s most prominent figures. The CEO of one of the leading labs has said publicly and repeatedly that AI could eliminate something like half of all entry-level white-collar jobs within a few years and push unemployment to ten or twenty percent. He didn’t say it triumphantly. He said it as a warning — that the industry and the government were "sugar-coating" what’s coming, and that the people producing the technology owe everyone the honesty of saying so.¹⁴

The warning is no longer hypothetical. The first large-scale evidence is in, and it lands exactly where the theory said it would: on the bottom rung. A Stanford study working from the payroll records of millions of workers found that since generative AI went mainstream, workers aged twenty-two to twenty-five in the occupations most exposed to it have seen employment fall about thirteen percent relative to everyone else — while older workers in the very same jobs held steady or grew. For young software developers the drop ran to roughly a fifth off the recent peak. The researchers titled the paper, with grim precision, Canaries in the Coal Mine.¹⁵ The supporting numbers rhyme: big-tech hiring of new graduates has fallen to around half its pre-pandemic level, and the unemployment rate for recent college graduates has drifted above the national rate — an inversion that is not supposed to happen.¹⁶ The entry-level economy, the on-ramp, the rung a person from the lower arm could use to climb onto the upper one, is being sawn off. The machine eats the bottom of the ladder first.

Here is the recursion. The same companies building the machine that removes the bottom rung are, in their own hiring, running the most selective competition for the top rung any industry has ever run. We are pulling up the ladder and holding an exclusive contest for the remaining seats, and calling both of those things meritocracy. The people hired into AI right now are, almost literally, being selected as the people who will stand on the upper arm of the world the industry is building. That is what I can’t un-see from inside the debrief. When I say I’d bet on her, I’m not staffing a job; I’m helping draft the roster for the side of the divide that gets a future — a divide my own work widens. The interview has quietly become a draft, and the thing it drafts you into is which arm of the K you spend your life on.

To be fair to the optimists — including the same lab leader, who has more recently leaned on an old argument: that cheaper cognition will create more demand for it,¹⁷ that the work eliminated will be exceeded by work created, that the lamplighters became electricians.¹⁸ Maybe. The long-run history is on their side, and I am not predicting permanent mass unemployment; I think that is the wrong fear. But notice that their argument is about aggregates — whether there will be enough total work — while mine is about distribution and selection. You can create a great deal of new work and still have a K, if the new work pools on the upper arm and the selection onto that arm runs through "fit" and "potential" and pedigree. The lump-of-labor fallacy is a real fallacy. It is also not a reply to this essay. Even in the rosiest aggregate future, someone is still deciding who gets onto the arm where the new work lives — and doing it with a rubric that mostly rewards already being there.

VA Sorting Machine That Knows What It Is

I am wary of how essays like this usually end, because I have written that ending before and been justly hit for it. The lazy landing is the sophisticated shrug: the system is vast, the forces are structural, nobody is to blame, isn’t it all terrible. That is not an argument but an aesthetic of helplessness — a way of enjoying the diagnosis while ducking the bill.¹⁹

So, plainly: the problem is not that hiring sorts. Sorting is unavoidable. Someone gets the offer and someone doesn’t; there is no version of a company, or a society, where every person is selected for every seat. A world without selection is not on the menu, and I am not nostalgic for one. The problem is sorting that does not know what it is doing — a selection machine that believes its own meritocratic story so completely it feels no obligation to the people it sorts out, because it has convinced itself it merely discovered their lesser worth rather than assigned it.

The difference between those two machines is the whole game, and it is the same difference I have spent years circling in other contexts: a system that knows it is a system, versus one that thinks it is a mirror. A hiring process that understands itself as a neutral measurement of desert owes nothing to the people it rejects, because it believes it has simply read a fact off of them. A hiring process that understands what it actually is — an actuarial bet about which arm of a diverging economy a person will land on, placed with a rubric contaminated by who already landed where — owes a great deal. It owes humility, because the winners did not earn the arm they were born on; they were dealt it and then handed a story that let them mistake the deal for an achievement. It owes the people on the other arm something other than the verdict that they deserve it. And for those of us inside the companies doing the widening, it owes the refusal of a specific pleasure: we do not get to enjoy the flattery of we only hire the best while building the thing that makes "the best" a smaller and more hereditary club. The honest posture is to hold both halves at once and let them hurt — I was selected, and selection is not the same as deserving, and the system that selected me is narrowing the gate behind me even as I praise it for its standards.

Naming this does not fix it, and I won’t pretend the naming is an action. It isn’t. The line will keep diverging whether or not I’ve described it well. But a sorting machine that knows it is a sorting machine is a different and less dangerous thing than one that thinks it is a mirror — because the first can be argued with, made to feel its obligations, asked to widen the gate it controls, and the second cannot, because it does not believe the gate is its doing. The first step toward a selection process that is accountable for what it selects is a selection process that admits what it is selecting. Which is futures. We are selecting futures and pretending we are measuring merit, and the pretending is the part that lets us off the hook.

I have three children. The youngest is only now old enough to play. And I notice — with a discomfort I haven’t resolved — that a great deal of what I am teaching them is, functionally, how to read as high-potential. How to present. How to be the kind of person a panel bets on. I tell myself this is preparation, and on my better days I believe it. But I know what the rubric rewards now, and I know which arm of the K it sorts toward, and on my worse days I cannot fully tell whether I am raising my children or training them to win an actuarial bet whose terms I find quietly monstrous. If we are no longer going to believe that anyone is watching from above and keeping the final score, then the only thing left to believe is that we are raising a better generation than the one that built the machine. I am trying to. But I would like, at minimum, to do it with the cover story switched off — to teach them to win the selection if they can, and never once mistake the winning for proof that they deserved to.

That is the most I have. Not a solution. A switched-off cover story. The room is sorting, and the least we can do is stop telling ourselves it’s a mirror — and say the colder, truer word for what the bet actually is.

Selection.