The Missing First Rung

A junior developer spends half a day tracing why an apparently harmless change broke something elsewhere. From a delivery dashboard, this can look inefficient. A senior colleague might have found the cause sooner, and an AI tool may soon do so faster still.

What the dashboard does not show is the junior learning that systems carry history, dependencies and local compromises. You acquire a feel for these things by doing the work, getting parts wrong and having somebody explain why.

This is why Stanford’s latest analysis of US payroll data deserves attention. It found no widespread, economy-wide job displacement through June 2026. Among 22- to 25-year-olds in highly AI-exposed occupations, however, employment stood 19% below where it would have been had it kept pace with less-exposed occupations. The difference came mainly from reduced hiring, while experienced workers showed no comparable gap.

The study does not prove that AI caused this. Its authors describe the findings as early indicators and point to several complications, including education, trends that predate generative AI and differences between their payroll sample and national surveys. Even with those caveats, the pattern exposes a weakness in many workforce plans.

Most organisations have asked junior work to do two jobs at once: produce today’s output and develop tomorrow’s experienced people. We did not have to distinguish between them while both happened in the same task. AI rather inconveniently makes the distinction visible.

A new joiner who mainly checks generated answers gets a different apprenticeship from someone who has traced an incident through a system, formed an argument from incomplete evidence or sat with a client’s objection. Recognising a plausible but dangerously wrong answer depends partly on having produced a few of those answers yourself and been corrected.

One company can hire fewer juniors now and recruit experienced people later. If enough employers make the same calculation, the pool they intend to recruit from becomes shallower. In a regulated business, an external hire can bring valuable experience, but not an immediate understanding of why this particular system, control or client arrangement works as it does.

The response is not to preserve routine work for nostalgic reasons. Junior colleagues should use AI. They should also remain responsible for their assumptions, tests and conclusions, with supervised exposure to incidents, client conversations, architecture decisions and regulatory trade-offs.

At the next workforce-planning round, I would ask what learning disappears with the tasks being automated, where junior colleagues can still make mistakes without causing serious harm, and whether experienced colleagues have time to explain the work rather than merely approve it.

The plan to ‘hire experience later’ may still work for a while. It relies on enough other organisations continuing to create it.

First published on LinkedIn on 15 September 2026.