Think about the analyst who joined your team three years ago. She built the quarterly forecast by hand and got part of it wrong. You called her in and spent twenty uncomfortable minutes asking where her renewal assumption came from. She didn’t have an answer. The next quarter, she did. Now she runs a small team and catches shaky numbers before they reach you.

Now say that forecast goes to an AI tool. The draft is ready before lunch, nobody sits in your office, and the cost line looks better. But you’ve lost those twenty minutes, and with them the conversation that turned an analyst into a future manager. Most companies have no line item for that.

People in leadership circles are talking about this. On October 12, at the IMF and World Bank meetings in Bangkok, Singapore’s president Tharman Shanmugaratnam warned that hiring is slowing at “the first rung in employment, your first job upon graduation.” A week later, a Forbes column asked a blunt question: if AI is taking entry-level jobs, who’s training your next CEO? If you manage managers or have any hand in succession planning, this one is yours.

What the Numbers Say

Start with Stanford. A Digital Economy Lab study by Erik Brynjolfsson, Bharat Chandar and Ruyu Chen used ADP payroll records covering millions of workers. Their first version, published in August 2025, found a 13% relative drop in employment for 22- to 25-year-olds in the jobs most exposed to generative AI, while older workers in the same jobs held steady. The August 2026 update, which runs through June, puts that young group about 19% below where it would have been if it had kept pace with peers in less exposed work. Most of the gap came from fewer hires, not more departures.

The Census Bureau sees the same thing from another angle. A September 2026 working paper followed graduates from the college majors most exposed to AI, led by computer science and software degrees. Once ChatGPT arrived, those grads were less likely to land a job right out of school, and their starting quarterly earnings fell about 13% compared with grads in less exposed fields. The researchers liken it to graduating into a big recession.

Job postings tell the same story. ZipRecruiter’s 2026 Graduate Report found entry-level roles made up 38.6% of postings at the start of this year, down from more than 44% three years earlier. And employers aren’t shy about the reason. Through September, U.S. employers cited AI in 120,136 announced job cuts, according to Challenger, Gray & Christmas, more than any other reason they gave.

The evidence isn’t unanimous. Economists Robert Fairlie and Jane Wu found no spike in unemployment among recent college grads this summer, and the Stanford team calls its results “descriptive patterns, not causal estimates.” Wharton’s Peter Cappelli has argued that some of what gets labeled AI displacement is plain cost-cutting with a trendier excuse. As a manager, though, you don’t need to settle that argument. Whether AI drove the slowdown or just gave employers cover for it, fewer people are walking in the front door.

The Deal Nobody Wrote Down

Entry-level jobs used to pay off twice. The company got the spreadsheet, the first draft of the memo, the meeting notes and the data cleanup. The new hire got the review. Economists Luis Garicano and Luis Rayo call this the old bargain: juniors paid for part of their own training by doing routine work. Nobody signed anything. The grunt work and the learning just lived inside the same task.

AI breaks that bundle. It’s good at exactly the work we used to hand to beginners, so the work goes away and the learning goes with it, unless someone deliberately moves the learning somewhere else. Garicano and Rayo’s model shows the career ladder holding under some conditions and training shrinking under others. Management choices decide which one you get.

Young workers already feel it. In a Cornerstone survey of 2,000 workers, 38% of Gen Z respondents say AI has fundamentally changed what their job requires, the highest rate of any generation. Yet 59% of Gen Z workers using AI on the job say their employer has never given them formal training on it. Your youngest people face the biggest change with the least help.

The starter tasks that don’t vanish tend to float upward. Someone still has to check the output, fix the edge cases and calm the angry client, and that someone is often a manager with a full plate. You save a junior salary and add hours to a senior one.

Why “Just Have Them Check the AI” Falls Short

The usual fix is to make juniors the reviewers: the AI drafts, the new hire grades. A World Economic Forum piece in June argued for exactly that shift, from doing the homework to grading it.

It sounds reasonable, but there’s a gap in the middle. Go back to the analyst. She learned what a bad assumption looks like because you found one in her own spreadsheet. It’s hard to check work you’ve never learned to do. Grading takes a standard, enough domain knowledge to sense when something is off, and the nerve to override a system that sounds confident. You get those from reps. A first-year employee who has only reviewed machine output may get good at spotting typos and still miss the logic problem underneath.

Review roles still have a place, but only alongside practice. Someone has to do enough of the real work, badly at first, with a senior person nearby, to build the instinct that makes their review worth anything.

The Bench Is Five Years Out

Hiring decisions stack up, which is why this belongs on a leadership agenda and not an HR wish list. The people you hire or skip this year become your first-line managers in three years and your director candidates in seven or eight. Thinning that group saves money this quarter. The gap shows up later, when you need someone to run a team and the person who would have been ready was never hired.

IBM’s chief HR officer, Nickle LaMoreaux, warns that cutting early-career hiring might save money now but leave a shortage of mid-career talent later. The Predictive Index’s 2027 workforce predictions make the same call, pointing to a leadership bench thinned out by disappearing entry-level jobs.

Most companies aren’t planning for it. Korn Ferry’s 2026 talent acquisition research found that 43% of organizations plan to replace roles with AI, with entry-level positions among the most affected. But only 22% factor AI readiness into succession planning. Put another way: more than four in ten plan to automate the bottom of the ladder, and only about two in ten account for AI when planning who climbs it.

Skipping the first rung has a quieter cost, too. Think about the best executives you’ve worked with. They probably have scars. They tried something early, it went sideways, and someone experienced helped them recover. Entry-level roles give people room to fail while the stakes are small. Remove them and your future leaders learn that lesson for the first time while running a department, with the stakes high.

What Some Companies Are Doing Instead

Not everyone is cutting. IBM said in February that it’s tripling its U.S. entry-level hiring this year. The jobs look different, though. Junior developers there still write code, now with AI’s help, and LaMoreaux said they spend “a lot more time talking to our marketing teams.” Whether the redesigned roles give beginners enough supervised practice to build judgment is something we won’t know for a while.

Brookings senior fellow Molly Kinder says to look to medicine. Residents make real decisions under supervision, while senior physicians coach them through cases, explain their reasoning and hand over more autonomy as it’s earned. Her pitch is a white-collar version, paid for by the companies that profit from AI. You don’t need a national program to borrow the idea, though. A senior person, a real decision and a debrief will do.

McKinsey describes another approach: employees take a first pass at selected work, then sit down with a manager to compare their reasoning against the AI’s. The gap between the two is the lesson.

Other companies go the opposite way. In April 2025, 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.” As an efficiency rule, that’s defensible. Without a counterweight, it can cut off the flow of future leaders. The risk is that nobody owns the other side of the equation.

What You Can Do Starting Monday

You probably don’t control your company’s hiring budget. You do control how work moves through your team, and that’s where most of this gets decided.

Before you automate any task, ask what a beginner learned from doing it and where that learning will happen now. If nobody can answer, you’ve found a training gap, not just an efficiency gain. Whoever approves the automation should name the manager who owns that learning, the same way you’d name an owner for any other risk.

Protect the review conversation. An AI-built spreadsheet can still get a good twenty-minute review. Ask junior people to defend the assumptions in the output as if they’d built it themselves, and to say where they’d push back. The tool saves two days of building. You keep the twenty minutes of learning.

Give early-career people a first pass on real work. Let them draft the recommendation before they see the AI’s version, then compare the two. Bring them into client calls and tough meetings with an experienced colleague, and debrief afterward. Short stretch assignments across teams build context faster than one seat does, and they cost little.

Change how you measure your managers. If first-line leaders are judged only on output, they’ll hand every beginner task to the tool. If part of their review covers how fast their people grow, the math changes. Developing people has to be somebody’s actual job, with time set aside for it. Otherwise it becomes nobody’s job.

Push succession planning further down. Most plans focus on who replaces the VP. Ask who replaces the people who’ll replace the VP, and where they come from if the bottom rung keeps shrinking. If the answer is “we’ll hire from outside,” ask where those other companies will get them.

Conclusion

AI automation is an easy sell, and a lot of the pitch holds up. The draft is done before lunch, the headcount request goes away, and the savings show up right away.

The bill comes later. The entry-level hire you skip in 2026 is the first-line manager you can’t promote in 2029 and the director slot you’re scrambling to fill in 2033. The Stanford and Census Bureau data and the job boards all show the first rung is already thinner in the fields AI touches most. The research can’t yet say how much of that AI caused, but the hole in your leadership pipeline looks the same either way.

You can fix this at the level where you work. You don’t have to win an argument about national labor policy. You have to keep a beginner in the room, keep the review conversation alive, and make someone accountable for what new people learn. Companies that do will have future leaders ready when they need them. The rest will be shopping for experienced people nobody trained.