The October “Who wants to be hired?” thread went up on Hacker News this morning: 76 points, 251 comments, and almost nothing in it but résumés from people who, this time last year, would have been fielding inbound. It sits beside two numbers the commentariat has decided are a cause and an effect. Stanford HAI’s 2026 AI Index has entry-level software developer employment down nearly 20 percent from its 2024 peak. Listings for “agentic AI” roles climbed more than 10,000 percent in twelve months. The story everyone tells is that the second number ate the first.

The story that fits the data is duller and worse. Nothing was eaten. It was sold.

The 10,854 Percent and the 20 Percent Are the Same Statistic

Dice’s 2026 tech jobs report — a snapshot pulled September 3, built on more than 7 million U.S. postings — and the Pragmatic Engineer’s survey of the market agree on the shape of things: “top” tech firms are hiring roughly 20 percent more software engineers than they were a year ago, U.S. and U.K. postings are up, Germany and France are down, and the three largest sources of openings are Apple, Amazon and IBM, not startups. Meta, which cut about 10 percent of its staff — some 8,000 jobs — in May and scrapped plans to fill 6,000 open roles, fell out of the top twenty entirely.

Read those together and the AI-displacement story stops working. If models were doing the junior work, total demand would shrink. Instead demand at the top is growing while demand at the bottom evaporates. That isn’t replacement. That’s reallocation — a firm deciding it would rather buy a finished engineer than make one. Layoffs.fyi counted almost 110,000 cuts across 137 tech companies so far in 2026, against roughly 125,000 in all of 2025. Companies aren’t shrinking. They’re changing what they’re willing to pay for. Note the other Stanford figure, the one nobody quotes: AI governance roles grew 17 percent. They’re hiring people to supervise the machines and nobody to inherit the codebase.

Atlassian’s CEO Said “the Mix of Skills.” Not “Less Work.”

Atlassian’s 1,600-person reduction came with a statement from CEO Mike Cannon-Brookes that AI changes “the mix of skills we need or the number of roles required in certain areas,” while insisting the approach is not “AI replaces people.” Notice the wording. Mix. Areas. It’s a sentence about composition, not volume — the most honest thing anyone has said about this labor market all year, and more honest than either the people insisting AI is coming for all of us or the ones insisting nothing is happening.

The old arrangement was never sentimental. A junior engineer cost real money for eighteen months and produced very little, and firms paid it anyway because they needed a mid-level bench by year five. That is capex with a four-year payback, and a four-year payback is invisible on a quarterly call. Cancel the program and the margin shows up now. The missing bench shows up in 2031, to somebody else’s management team, in a division that hasn’t been spun out yet.

One recruiter at a mid-size cloud vendor described the whole thing in a Slack message to a colleague on her own team: two junior requisitions approved in 2024, denied in 2025, refiled this spring as a single senior opening. Same budget. Fewer names.

The Bill Arrives Around 2031

Here is the testable part. If the reallocation thesis is right, the next four years produce a barbell: a deep bench of very senior people, a large and growing cohort that never got a first job at all, and a hollow middle. Then the mid-level shortage bites — not everywhere, and not dramatically, but in exactly the places that cannot simply bid up senior salaries without wrecking their own cost structure. The firms currently congratulating themselves on margin will discover that you cannot promote from within a group you never hired.

If I’m wrong, the correction is easy to state and easy to watch. Junior postings come back on their own. The Stanford entry-level figure recovers past its 2024 peak without anyone changing anything. The “agentic AI” listings stop being a substitute for a hiring class and start being a complement to it — a tool that makes a first-year engineer worth paying, which is what every honest vendor pitch already claims.

Who Has to Update Their Priors

Both camps, and that’s the point. The people who say AI is stealing the jobs have been handed an alibi by management and accepted it gratefully, which means they’re now arguing for the wrong remedy against the wrong cause. The people who say the market is simply working — that the degree was overpriced, that the kids should retrain — have to explain why a market that is working so well has stopped producing a first rung. Markets are good at prices. They are not good at funding things whose payoff belongs to a successor.

And the executives who put “AI” in the memo should have to say out loud what they actually decided: that training junior engineers is a cost they no longer intend to carry. That’s a legitimate decision. It’s defensible, it’s legal, and someone with a spreadsheet probably has the NPV to prove it. It is just not a technology story, and it should not get to hide behind one.

Watch the résumé thread. It will be back on the first of next month, and it will be longer.

Sources