Cutting entry-level jobs to save money could cost companies future managers

Matt Garman doesn't think replacing junior developers with AI is smart business. Amazon $AMZN's cloud chief called the idea "one of the dumbest things I've ever heard." His position explains why Amazon Web Services is hiring 11,000 interns and recent graduates this year, even after the company cut 14,000 corporate jobs last fall.

That's the shape of a bet playing out across corporate America right now. Companies are cutting entry-level jobs and rehiring some of them in the same year, showing just how hard it is to both automate the tasks junior employees used to do and protect the pipeline that typically produces senior talent. If companies let AI eliminate the entry-level rung, they could lose the mechanism that turns 20-something up-and-comers into top-tier managers.

The data says that bet is already straining. Young workers just starting their careers are concentrated in the routine tasks AI already handles well. That group has been losing ground steadily since 2022, and the erosion has widened every month for close to four years. What's less clear is whether the jobs disappearing are the ones companies can safely automate, or the ones they'll miss the moment they need someone who's actually done the work before.

Widening employment split by career stage

Erik Brynjolfsson, a Stanford economist who runs the university's Digital Economy Lab, wanted to know whether anecdotes about the junior tech job implosion held up against real payroll data. His team pulled millions of ADP records, and after adjusting for shocks specific to individual companies, it found that workers between 22 and 25 in the jobs AI reaches first had lost 16% of their employment relative to older colleagues doing the same work.

Among software developers, employees aged 22 to 25 lost nearly a fifth of their jobs relative to the late-2022 peak, while developers over 35 kept growing through the same stretch. Young marketing and sales managers lost jobs, too, but on a smaller scale. Stock clerks fared about the same regardless of age. Young health aides actually gained jobs faster than their older colleagues did. The losses track exactly which tasks AI has gotten good at and which jobs it can't fully replace.

Brynjolfsson hasn't stopped watching. He teamed up with ADP Research to build a live dashboard tracking the same cohorts through April, and the decline for the most exposed 22-to-25 group has sharpened to nearly 4% a year, up from under 3% after the first 12 months. The monthly numbers bounce around, but the yearly trend has run in one direction for more than three years straight.

Automation versus augmentation in practice

AI automates some tasks and augments others, and that distinction is what decides who loses ground and who doesn't. Occupations where AI mostly replaces what a person used to do show the steepest losses among young workers. Occupations where AI helps a person do more without eliminating the role show stable or growing employment for the same age group.

Entry-level work sits overwhelmingly in the first category. Most of a junior job is drafting a first version of a report, reconciling routine transactions, or summarizing a call. That's exactly the material AI has gotten good at fast. The tacit half is different. It's the idiosyncratic shortcuts and workarounds a person picks up on the job that never make it into a training manual or a company database. That kind of knowledge builds up over years, and if a model can attain it in the first place, older industry insiders become even more valuable to the workforce, not less.

Companies are also hiring fewer junior people rather than cutting the pay of the ones they keep. So the person affected most, at least financially, is the young applicant who can't get through the door. The only ones who don't have to worry about this tradeoff are the senior managers who've already gone through the early stages of career building.

What entry-level jobs build for companies

Professors Amy Edmondson and Tomas Chamorro-Premuzic argue that cutting these jobs trades a small, immediate saving for something companies notice only once it's gone: the pipeline that turns a 24-year-old into a manager who has actually done the work. Every capable executive picked up judgment somewhere unglamorous, answering an angry customer, running a shift that went sideways, and catching a mistake before it reached their boss. Remove the jobs where that happens and the next generation of managers arrives without knowing what it takes to actually do a good job.

The same jobs surface problems nobody assigned anyone to look for. Junior employees sit closer to the friction points than anyone above them, and they tend to notice what's broken before management does. Take that layer out, and a company loses a source of correction it didn't realize it depended on.

Work gives people more than a paycheck. It offers purpose and a place to belong, and an entry-level role is usually where someone finds that for the first time. Cutting these roles at scale doesn't just reshape a hiring plan. It changes what a large number of 22-year-olds do with their time, and history hasn't been kind to economies that leave large numbers of young adults without anywhere meaningful to go.

Redesigning the job instead of cutting it

Garman's answer is to keep the role and change what fills it. Edmondson and Chamorro-Premuzic point to accounting firms doing something similar. AI clears the routine matching and reporting work, and junior staff spend their freed-up time chasing anomalies and talking through problems with clients, the work a machine still can't do convincingly.

A similar shift shows up wherever a company bothers to redesign the role instead of just automating it. Job descriptions become less about performing a task and more about checking whether the AI did it correctly, from producing a first draft to catching what the draft got wrong. That kind of quality control still requires employees to know what looks good and catch anything that looks off.

That's a universal skill. As AI works its way into higher-level decisions, the ability to catch it being confidently wrong becomes something every level of a company needs, not just at the entry level. The only place anyone learns it is by doing the junior work some companies are busy eliminating.