AI can speed up entry-level work. Cut the training those assignments gave, and the leadership bench thins out.

Picture an entry-level analyst whose forecast just missed. The spreadsheet took her two days. The 20-minute review is where she learns. Her manager walks through each assumption, stops at the one about customer renewals and asks where it came from. She doesn’t have a good answer, and now she knows to look for one next time.

Now give the spreadsheet to an AI tool, and the draft is ready before lunch. If her manager stops asking her to defend the assumptions inside it, the lesson goes too. The savings show up right away. The missing experience can show up years later, when the company needs her to run a team.

Singapore’s president, Tharman Shanmugaratnam, raised the same worry at the IMF and World Bank meetings in Bangkok on October 12. Hiring is slowing at “the first rung in employment, your first job upon graduation,” he said. Governments should work with employers so graduates “get that first foothold in the market, and they can learn from there.”

What The Entry-Level Hiring Data Shows

In the fields AI touches most, the first rung is getting harder to reach, though the wider evidence is mixed. In a September working paper, Census Bureau researchers followed graduates of the college majors most exposed to AI, a group led by computer science and other software degrees. After ChatGPT arrived, those graduates became less likely to land a job right out of school, and their initial quarterly earnings fell about 13% relative to graduates in less exposed fields. The researchers liken that to graduating into a large recession, and the gap narrows over time. About half the earnings drop reflects a shift toward lower-paying industries such as retail and food service.

Stanford’s Digital Economy Lab found something similar in its August analysis of ADP payroll data through June 2026. Employment for 22-to-25-year-olds in the most AI-exposed occupations was about 19% lower than it would have been had it kept pace with peers in less exposed work. Most of that gap came from fewer hires rather than more departures.

Companies are also saying it out loud. Employers have cited AI in 120,136 announced U.S. job cuts so far this year, more than any other reason. Some openings never get as far as a posting. Last year, Shopify CEO Tobi Lütke told his teams that before asking for more headcount , they “must demonstrate why they cannot get what they want done using AI.”

Not every study finds the same pattern. Robert Fairlie and Jane Wu found no spike in unemployment among recent college graduates this summer. That’s worth taking seriously, though an unemployment rate has limits: a graduate stocking shelves counts as employed, whatever their degree. What the research can’t yet say is how much of the hiring slowdown AI caused. Stanford calls its results “descriptive patterns, not causal estimates,” and its gap narrows once education is taken into account.

Entry-Level Jobs Paid For Themselves Twice

Economists Luis Garicano and Luis Rayo describe the old bargain: juniors paid for part of their training by doing the routine work. The firm got the spreadsheet, and the analyst got the review. Their theoretical model asks what happens as AI takes on more of that routine work. Under some conditions the career path still pays for itself, and under others the training shrinks. It depends on how much more an expert is worth than a beginner who knows just enough to beat the AI.

One proposed fix is to have juniors review the AI’s output. That sounds sensible until you ask how a beginner learns what a bad assumption looks like. Remember the analyst? She learned what a bad assumption looks like because her manager found one in her own spreadsheet. It’s hard to check work you’ve never learned to do.

This is where it becomes a leadership problem. The analyst who learns to spot a bad assumption becomes the manager who decides which assumptions her team can live with, and maybe one day the CEO who has to question an analysis she didn’t build. IBM’s HR chief, Nickle LaMoreaux, has warned that cutting early-career hiring might save money in the short term but create a shortage of mid-career talent later. In my experience, the best executives are the ones with scars from climbing the corporate ladder. They tried new things, some didn’t work out, and they had to recover from the failure. Early assignments can give people room to make those mistakes while someone experienced is there to catch them.

Someone Has To Own The Training

Medical residency is the established model of supervised practice. Residents make real decisions under supervision while senior physicians “coach them through cases, explain their reasoning, and gradually grant autonomy,” and Medicare helps cover hospitals’ training costs. Brookings senior fellow Molly Kinder wants something similar for white-collar work, paid for by the companies that profit from AI rather than by taxpayers. Tharman would bring governments in, too.

IBM is redesigning its own entry-level roles. LaMoreaux told a Charter summit in February that IBM is tripling entry-level hiring . Junior developers there still code, now with AI’s help, and spend “a lot more time talking to our marketing teams,” she said. The test is whether those redesigned jobs give juniors enough supervised practice to build judgment. The announcement describes an approach, not results.

The AI risks I worry about most are the ones without a clear owner , and training can easily become one of them. Before a company automates an entry-level task, someone should be able to say what a beginner learned from it and where that learning happens next. McKinsey describes one approach: let employees try selected work first , then go over the differences between their reasoning and the AI’s with a manager. Another is bringing juniors into client meetings with an experienced colleague and talking through the decisions afterward.

The executive who approves the automation should name the manager responsible for that learning. An AI-built spreadsheet can still get a good review meeting. The company just has to keep a beginner in the room.