The Class of 2031

A short story about a college freshman who started in 2027, and then what I think it means for anyone deciding whether to hire juniors next year.

---

Nadia's advisor tried to talk her out of it.

It was the fall of 2027. Computer science enrollment at her school had dropped by about a third in two years. Entry-level postings were down almost everywhere. A girl from her high school had graduated the spring before and was working at a rental car counter.

"Pick something the machines can't do," the advisor said.

She picked computer science anyway, and added statistics alongside it.

The statistics part had nothing to do with the job market. When she was sixteen she'd spent most of a summer arguing with her father about whether a player on the White Sox was actually good. She had the counting stats. He had something he'd read. She lost the argument in front of her uncle and two cousins, and what bothered her for the next year wasn't losing. It was that she couldn't tell whether she'd been right.

So she took the stats sequence, and kept taking it after the requirement was satisfied.

She made a rule for herself that first year. She'd use AI on everything, in every class, as hard as she could. She'd never turn in anything she couldn't defend out loud.

Sophomore year she asked for a regression on county housing data and got back something clean. The coefficients were tidy. The write-up was persuasive. She spent an hour with it and found the model had no control for the thing driving most of the variation. The result wasn't wrong. It just didn't mean what it said it meant.

She would not have caught that as a freshman.

She'd also started volunteering at the campus food pantry that year, Saturday mornings, because a friend needed someone to go with her and then stopped going. Denise ran the place. Denise had run it for eleven years and scheduled ninety volunteers on a whiteboard and a legal pad, and complained about it constantly.

Junior year Nadia offered to build her something. It took a couple of weekends. Most of it she wrote with an assistant and checked. The date handling she didn't check. It looked fine and she had a midterm.

For about three weeks it double-booked shifts across the daylight saving change. Twice the pantry opened with nobody there.

Denise was nice about it on the phone, which somehow made it worse.

Nadia knew how to check that code. She'd checked harder things that semester. She just hadn't thought to, and she spent a while afterward trying to work out what would make her think to next time. She never fully solved that. She got better at it.

Columbus, summer 2030

By that summer she was interning in the analytics group at a food distributor outside Columbus. Family-owned, about a thousand people, and nobody there would have called it a technology company.

Nobody assigned her anything important. She got the leftovers, which is what interns get. Two and a half years of carrier invoices nobody had reconciled. Somebody had started 2028 and given up in March.

She reconciled them. Then she kept going, mostly because something was bothering her.

What the company was paying on a handful of lanes didn't match what the contracts said it should be paying. Not dramatically. Enough to notice.

She built a model. Then she spent longer trying to break it than she'd spent building it, because the first version told her something too good, and she'd learned what that usually means. She rewrote it twice. The second rewrite was the one that held.

She showed it to her manager in a meeting she'd asked for and that got pushed twice. Her manager showed the CFO. Some of it ended up in the material that went to the board that fall, inside a conversation about freight spend that was already on the agenda.

Nobody had planned for that.

She'd been given the invoices because they were the least important thing on the list.

Ellis

Ellis runs engineering and analytics there. He hasn't decided what to do next year and it's getting late to decide.

He cut the new-grad program in 2028. The math was obvious at the time and he'd run it the same way again. A junior cost him a salary plus two senior engineers' attention, and the work juniors used to do had become the work his tools did before lunch.

He's never said the other part out loud, including to himself with much clarity. He thinks the people coming out of school now are softer than the ones he hired in 2019. He thinks they've had the hard parts done for them.

He has a fair amount of evidence for this. He's careful about who he says it to.

Nadia bothered him a little. Not in a way he could do anything with. She was one intern, and he'd watched her for ten weeks from a distance, and she'd done something with a stack of invoices he wouldn't have expected from a second-year analyst.

What he didn't price into the 2028 decision was what would happen to the seniors.

They spend their days now reviewing output they didn't write. There's nobody below them to hand the small decisions to, so every decision stays with them. Two of his best people have told him some version of the same thing this year. The job stopped being the job.

He has budget for two senior hires or six new grads. Not both.

The seniors will be productive in six weeks. That's not a small thing when he's already short. The grads won't be useful for a year, maybe longer, and he'd be spending his seniors' remaining patience to train them.

The other side of it is that he can't buy a 2036 senior engineer in 2036. He can only hire one now and wait.

He also knows every company he competes with ran the same arithmetic he did, and most of them landed where he landed. That makes the grads cheap. It doesn't make them easier to justify to a CFO looking at the same quarter he is.

If he's wrong, he won't find out for five years. By then it won't be a decision he can revisit.

That's the number he keeps coming back to. Not the budget.

The five.

What I think this says

Here's the claim, stated plainly so you can disagree with it.

The entry-level rung didn't disappear. It moved earlier, and it stopped being something employers build.

Apprenticeship has always been how judgment gets transmitted, because judgment can't be taught from a syllabus. You watch someone make the call, you make it yourself badly, someone corrects you. For most of the last forty years, companies owned that process. The first two years of a career were the apprenticeship, paid for by the employer, funded by work that was real but low-stakes enough to survive a novice.

That work is what the tools took first. So companies stopped paying for the apprenticeship, which was rational in every individual quarter and ruinous in aggregate.

What Nadia has isn't a tool advantage. Every one of her classmates has the same tools. What she has is four years of failing with them somewhere failure was free, plus a discipline rigorous enough to tell a sound result from a confident one.

She arrived already apprenticed. Nobody at the company paid for it, and nobody there could see it on her transcript.

The part I'd want a leader to sit with is that this changes what the hiring problem actually is. Ellis frames his as a budget question: two seniors or six grads. But the real constraint is that he has no way to tell which of the six are Nadia. His filters were built for a world where the credential carried the signal, and the signal moved to things his process doesn't look at. Ten weeks with an unwanted stack of invoices told him more than four rounds of interviews would have.

Here's what would make this wrong. If people like Nadia are rare, one in twenty rather than one in four, then Ellis's arithmetic holds and the six-grad bet fails no matter how good his recognition gets. I don't know the ratio. Nobody does yet. It's the number that decides this, and it isn't measurable until the companies making the bet are already five years into it.

What I do know is that most organizations are answering this by not answering it. The junior program doesn't get cancelled in a meeting. It goes unfilled, quarter after quarter, because each individual quarter the arithmetic says wait.