The number that made the room happy was the one nobody studied for very long. An operations chief I spoke with this spring had cut his incoming analyst class by about half and felt good about it, because the tools his team had bought could draft the models and clean the data those analysts used to grind through by hand. The year's math worked. Fewer salaries against the same output, and a margin line that finally moved the direction his board kept asking for. What he had not priced, because it lands in no quarter he will ever be measured on, was where the senior analysts of 2033 were now supposed to come from.

He is not an outlier. The pullback in early-career hiring is broad enough that it already shows up in payroll records. A team at the Stanford Digital Economy Lab went through ADP data covering millions of workers and found that employment for people aged 22 to 25 in the occupations most exposed to AI, software development and customer service among them, has fallen about 13 percent relative to everyone else since generative tools came into wide use.1 Their read on it is plain enough to sting. As one of the researchers put it, what younger workers know overlaps with what the models can already replace.1

The job you automated was also a classroom

Here is the part the margin math leaves out. The entry-level grind was never only output. It was the mechanism by which a person turned into someone worth keeping. The analyst who spends two years building the models by hand is the one who later catches when a model is wrong, because she remembers every place she once broke one. Andrew McAfee, who studies this at MIT, put the question better than I can. How else are people going to learn to do the job, he asked, except through on-the-job learning and apprenticeship.2 Pull back on entry-level hiring, he warned, and you are sacrificing the skilled people of the future.2 Automate the grind and you keep this year's output while closing the school that produced everyone standing above it.

Every junior role you delete is a unit of output saved this year and a senior you will not have in ten. Only the first one shows up on the dashboard.

Why the bill stays invisible until it is too large to pay

The trade is attractive precisely because its two sides arrive on different clocks. The saving is immediate, legible, and yours to claim in the review where it counts. The cost shows up years later as a thinning layer of mid-level people who can actually run things, by which point the person who made the cut has been promoted on the strength of the margin or has moved on entirely. Nobody's name sits on the 2033 bench. No quarterly metric tracks it. So the decision gets made over and over by people who will never personally meet its consequence, which is about the cleanest way I know to guarantee a bad call keeps getting made.

The companies betting the other way are telling on the rest

Not everyone is making the cut, and the exceptions are worth reading. IBM has tripled its entry-level hiring. Amazon has held its plan to bring on eleven thousand software engineering interns. Salesforce is taking on a thousand new graduates.2 These are not sentimental places. They have decided the pipeline is worth more than the short-term saving, and they grasped the move the others missed. You do not keep hiring juniors to do the grind the tools now handle. You rebuild the entry role around the work the tools cannot do yet, the judgment and context that only come from doing the job, so the apprenticeship still happens while the busywork goes to the machine. That is harder than zeroing a line in the hiring plan. It is also the difference between using AI to compound your people and using it to eat your own future.

So before the next planning cycle zeroes out the entry-level line because the tools can cover it, there is a question worth forcing into the room. Who here owns the answer to where our senior people come from in ten years, and can they give it without waving at a hiring market everyone else is busy hollowing out at the same time? If no one owns that question, the saving on this year's slide is a loan against a bill someone else will be standing there to pay. What would it take to redesign one entry-level role in your shop around what the work will actually need?