AI is quietly killing the junior job. Where do your future leaders come from?

Nobody's firing the graduates. Companies are just hiring fewer of them, and that quieter change could leave a hole in your leadership pipeline in five years' time.

The latest evidence comes from Stanford economists, in the August 2026 update of their paper "Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence." As Ars Technica reported, employment for workers aged 22 to 25 in the most AI-exposed occupations now sits 19% below their peers in less exposed fields. A year ago that gap was 13%.

The researchers used a large sample of anonymised payroll data from ADP. Across the whole economy, they found little difference in employment between the jobs most and least affected by AI. The damage is concentrated at the bottom of the ladder.

The on-ramp is closing

Here's the detail that should worry anyone who owns workforce planning. Since 2022, employment for 22 to 25 year olds in the 40% of jobs most affected by AI has fallen by about 11%. In the other 60% of jobs, employment for the same age group grew by 10%.

And it's happening through hiring. The researchers found the effect shows up mainly as lower hiring rates for entry-level workers in AI-affected fields, rather than more firings or more people quitting. It also shows up in fewer jobs rather than lower pay.

That makes it easy to miss. There's no layoff announcement and no headline. A team that used to take on four graduates a year takes on two, then one, because the tools now handle the first-draft work juniors used to do. Each decision looks sensible on its own.

Erik Brynjolfsson, who leads the research, told The Washington Post he's more worried than he was about "a labor market that keeps its overall employment level while quietly closing the on-ramp for people starting their careers." That's the risk in one line.

Which jobs are exposed

Not every job touched by AI is hit the same way. The researchers split AI use into two kinds, drawing on Anthropic's Economic Index. Automative use replaces work a person used to do. Augmentative use helps a person do their job better.

Jobs where automative use is common showed the worst entry-level employment trends. Accountants and auditors, and receptionists and information clerks, were among the roles judged most exposed to automation. Chief executives and registered nurses were among those using AI mostly to augment their work.

"The findings are consistent with automation-oriented uses of AI substituting for labor while complementary uses are associated with flat or rising employment," the researchers write.

For a leader, that's a useful lens. Look at the entry-level roles in your own business and ask which ones are mostly first-draft work that a tool now does. Those are the roles quietly disappearing from your hiring plan.

Why the junior job mattered

Junior roles have always been where people learned the job. Think about the graduate accountant reconciling numbers, or the junior analyst building a first model that a manager tears apart. The work itself was often dull. The learning came from doing it badly and getting feedback.

The Stanford team gets at this with a distinction between two kinds of knowledge. Codified knowledge is the formal, documented kind you can learn from textbooks or written procedures. Tacit knowledge is, in their words, "acquired through practice, mentorship, and repeated exposure to real situations."

They found occupations heavy in codified knowledge had slower entry-level employment growth. Occupations heavy in tacit knowledge had faster employment growth for mid-career and senior workers.

Put those two findings together and you get the problem. AI is good at the codified work juniors used to do. That codified work was how juniors got the reps that built their tacit knowledge. The senior people you'll need in 2031 are supposed to be learning right now, and in a lot of companies there's no seat for them.

The bill comes later

Cutting graduate intake makes this year's numbers look better. The cost shows up years later, when you go looking for an experienced manager who knows your business and find you never hired the people who'd have grown into the role.

You can try to buy that talent from outside. So will every other company that made the same call, and they'll all be bidding for a smaller pool. I'd rather not bet the leadership bench on winning that auction.

There's also a culture cost. Junior people ask naive questions that expose lazy assumptions. They bring new habits in, including habits with AI tools that older staff often lack. A team made up entirely of experienced people using AI can be efficient and very hard to change.

Signs it's already happening to you

Most companies doing this haven't decided to do it. It happens through a series of small, reasonable calls, so it's worth looking for the pattern on purpose.

Start with the requisitions. When a junior role opens up after someone leaves, does it get backfilled, or does the team say they'll manage with the tools for now? If the answer is usually the second one, your entry-level headcount is shrinking without anyone signing off on it.

Look at who's doing the first drafts. If senior people are now prompting a tool for work they used to delegate, they've saved some time and the junior has lost a learning opportunity. That trade might be fine for a while. Over a few years it adds up.

Then check your graduate program. Plenty of programs are still running, with fewer places each year and the same rotations they had before AI. If the rotations are full of tasks a tool now does in seconds, the graduates are learning a job that won't exist by the time they're promoted.

Finally, listen to your mid-level managers. They're often the first to notice when the people below them aren't getting enough real work to grow. If they're saying juniors seem less ready than they used to, take it seriously. It could be the tools, and it could be how you've redesigned the work around them.

What I'd do instead

Keep hiring juniors, but change what they do. If AI now handles the first draft, make the junior's job reviewing it. Have them find where an AI-generated reconciliation is wrong and explain why. That's a faster way to build judgement than doing the draft by hand, as long as someone senior checks their checking.

Redesign the apprenticeship on purpose. Most companies never wrote down how juniors learned, because it happened by default. Now it won't. Map the tacit skills your mid-level people need, then work backwards to the experiences that build them, and make sure your entry-level roles still include those experiences.

Pair juniors with seniors on real problems. Mentorship was always part of how tacit knowledge moved. With fewer juniors and busier seniors, it won't happen unless someone schedules it. Put it in the plan and in the performance goals of the people who are meant to do it.

Train juniors in the tools from day one. A graduate who can use AI well and has the judgement to spot when it's wrong is worth more than one who can do the old manual task. Make that judgement the core of your graduate program. An AI training session built around real work is a fair place to start, as long as it teaches people to check the tool as well as use it.

And measure it. Track your entry-level intake and your internal promotion rate over the next few years. If both are falling, you're running down your future leadership bench, whatever the efficiency numbers say.

The question for your board

Across the whole economy, the Stanford data shows little difference between the most and least exposed jobs. The trouble sits in a narrowing door at the start of careers, in exactly the fields where AI takes over routine work.

That's a choice companies are making one hiring round at a time, often without noticing. So put it on the agenda. Ask how many graduates you hired this year compared with three years ago, and what those roles now involve. Then ask who, specifically, will be running your teams a decade from now, and where they're going to learn how.