Thinking

The missing first rung: who trains the next senior?

If nobody starts on the first rung, who becomes the senior who can judge what AI produces?

Every senior was once a junior. They learned by doing the work nobody else wanted: the first draft, the first pass at the research, the routine code, the meeting notes. Someone more experienced checked it and sent it back, and over a few years the junior became the person doing the checking.

That entry-level work is exactly what AI now does best. So companies are hiring fewer juniors, and the data shows it.

In the short run, that’s a saving. In the long run, it raises a question few companies are asking: if nobody starts on the first rung, who becomes the senior who can judge what AI produces?

The first rung is thinning

The clearest evidence comes from payroll records. Economists at Stanford’s Digital Economy Lab, led by Erik Brynjolfsson, track millions of US workers through ADP, the payroll company. By June 2026, employment of 22- to 25-year-olds in the jobs most exposed to AI was 19% below where it would have been had it kept pace with less-exposed peers. Experienced workers in the same jobs showed no such gap, and the divergence has widened since the team first reported it in August 2025.

Two details matter. The drop comes mostly from fewer hires, not layoffs: the juniors already inside are staying, but fewer new ones are let in. And it sits in jobs where AI substitutes for the work. Where AI mostly helps people do their work, employment is flat or rising.

A Harvard study of 62 million workers at 285,000 US firms points the same way. From early 2023, junior employment at companies adopting generative AI fell 7.7% relative to similar companies within six quarters, through slower hiring. Senior employment kept growing.

The pattern shows up in single industries too. In tech, new graduates made up 7% of hires at the 15 largest companies in 2024, according to the venture firm SignalFire, and new-graduate hiring there was down more than 50% from 2019. In UK accounting, the Big Four all took on fewer graduates.

The UK Big Four took on fewer graduates. KPMG cut its intake by almost a third.

Graduate intake, 2023 to the following year: 6,499 to 5,442 across the four firms

Source: The Telegraph, via Scottish Financial News, June 2025. Redrawn by Gruow.

AI isn’t the only cause. Higher interest rates, over-hiring during the pandemic and offshoring all play a part, and none of the Big Four firms tied their cuts to AI in their own words. But the payroll data compares exposed and less-exposed jobs over the same years, so those wider forces affect both sides. The gap that’s left lines up with AI.

Why the first rung mattered

Junior work was never only cheap labour. It was how expertise got passed on. Much of what seniors know can’t be written down: which details matter, what a client will push back on, when a number looks wrong. People pick that up by doing the work, getting it checked, and doing it again.

Surgery has already run this experiment. Matt Beane, a researcher now at the University of California, Santa Barbara, spent more than two years watching hundreds of robotic and traditional operations at five hospitals, and interviewed surgeons and trainees at 13 more. In open surgery, the senior surgeon needs a resident’s hands throughout, so residents learn by doing. With a surgical robot, the senior surgeon can operate alone. Residents were left clearing smoke at the bedside or watching from a second console, waiting for a turn.

Only a minority found a way through. They practised on simulators and recorded operations, and pushed themselves on hard work with little supervision, mostly out of sight. Beane called this “shadow learning”. Most residents struggled, and all of them were still certified to operate.

That’s the risk now in offices. When a tool lets the expert work alone, the novice loses their place in the work, and nobody notices, because the work still gets done. The cost shows up years later, when the people who should be ready to judge the work never had the chance to learn it.

The Harvard study found one more sign of this. At firms adopting AI, promotions of juniors into senior roles rose, while fewer new juniors came in behind them. The ladder isn’t disappearing. It’s getting shorter at the bottom.

AI can also be the best teacher juniors have had

The same technology that removes junior work can speed up junior learning, when it’s built for that.

In the study of 5,179 customer support agents by Brynjolfsson, Danielle Li and Lindsey Raymond, the AI assistant was trained on the conversations of the company’s best agents. It raised the output of new and less-skilled agents by 34%. Agents with two months on the job who used it performed as well as agents with six months who didn’t. In the researchers’ words, the tool spread “the potentially tacit knowledge of more able workers” to everyone else.

The Harvard and BCG experiment with consultants found the same tilt. Consultants who scored below the median before the study improved 43% with AI; those above it improved 17%.

So AI isn’t the problem for juniors in itself. Used well, it compresses years of learning into months. The problem is a company that uses AI to skip hiring juniors at all, and loses the people it would have taught.

Rebuilding the first rung

The answer isn’t to ban AI from junior work. It’s to redesign that work so people still learn from it.

  • Keep hiring juniors, on purpose. Maybe fewer, but planned rather than left to chance. With AI, a new hire can reach useful work in months instead of years, so each junior costs less to train than before.
  • Make checking the new apprenticeship. If AI writes the first draft, the junior’s job is to review it alongside a senior who explains what’s wrong and why. That’s where judgment gets built.
  • Turn your best people’s know-how into material. The support tool worked because it learned from the top agents. Record decisions, keep examples of good work, and explain the reasoning, so both juniors and AI tools can learn from it.
  • Leave room to struggle. Beane’s successful residents practised hard work just past their ability. Give juniors some tasks to do without AI, with someone to check them.
  • Count it as a cost. A company that stops hiring juniors is borrowing against its future seniors. Put that in the same plan as the savings.

When we build tools that take over routine work, we ask who used to learn from that work, and where they’ll learn now.

AI can do much of a junior’s work. It can’t become your next senior. Someone still has to.

Sources