# AI and junior jobs: what happens to people starting out now Author: Quadra URL: https://quadra-website.matt-madd.chatgpt.site/risorse/articoli/ai-lavoro-junior Published: 2026-09-18 Updated: 2026-09-18 Language: en Reading time: 5 min For decades, people learned a profession by doing the tedious work. That work is disappearing first. ## At a glance - If AI takes over the tasks through which people learned a profession, businesses need to rethink how junior staff develop. - Employment data signals change, but separating AI effects from economic factors requires care. - The useful question is not only how much work to automate, but how to preserve learning, judgement and experience. A trainee lawyer spent their first two years reading contracts and looking up precedents. A junior designer prepared production files. A new hire in accounts reconciled invoices. Nobody thought that work was valuable in itself. It was tedious and repetitive, which was precisely why it went to the newest person. But while doing it, they learned things nobody would ever have explained aloud: what a badly written contract looks like, why that client always disputes the same line, which supplier gets the amounts wrong. Tedious work was not only a way to produce. It was a way to learn. ## What the data says Stanford's study of entry-level employment, updated in August 2026, tracks the issue using an unusually solid source: payroll records for millions of American workers, through June 2026. The finding has two sides. There is no sign of mass displacement across the economy. But employment among 22–25-year-olds in the occupations most exposed to AI is around 19% below the level it would have reached had it grown in line with their peers in less exposed occupations. Experienced workers show no comparable gap. In absolute terms, between November 2022 and June 2026, employment among 22–25-year-olds in the most exposed occupations fell by around 11%, while in the least exposed it grew by around 10%. **AI is not firing young people: it is reducing junior hiring.** Almost all of the gap comes from hiring, not an increase in layoffs. The interpretation is not unanimous. Some economists argue that the real cause is the sharpest interest-rate tightening cycle in forty years, and that the timing does not add up: six months after ChatGPT's release, companies could not already have redesigned their processes and changed their staffing. The Stanford authors have tested those objections: they excluded the entire technology sector, isolated the effect of remote work, and examined the interest-rate hypothesis, which points in the opposite direction, since the most rate-sensitive occupations are also the least exposed to AI. The gap did not close, and it keeps widening. It is not definitive proof. It is a signal that has persisted for four years. ## What changes for a business Here, the topic stops being a conference debate and becomes a practical issue, even for a small business. If repetitive tasks are automated, the question is not only how many people are needed. It is: where will they learn? A company that automates supplier invoice checks saves time, and rightly so. But those checks were also how a new accounts employee learned how purchasing worked, who ordered what and where mistakes arose. Remove them, and you are left with someone who knows how to use a tool but cannot recognise when it is wrong. ## Something manufacturing has already learned This problem is not new. Manufacturing businesses faced it twenty years ago, when computer-controlled machines replaced the manual work through which people learned the trade. The older turner had learned on the shop floor, making mistakes on parts. The new operator learned to load a program, and when the machine produced an out-of-tolerance part, they did not know where to begin. What worked, where it worked, was not going backwards. It was deliberately rebuilding what had previously happened on its own: structured mentoring instead of chance encounters, shop-floor instructions written with those who knew the trade, and practice on scrap material where mistakes cost nothing. The difference between companies that handled that transition well and those caught off guard was not the technology, which was the same for everyone. It was how early they asked themselves the question. It is worth remembering, because the next wave is coming much faster and affects office work, where nobody has ever had shop-floor instructions. ## Where we are heading The direction is not that young people will no longer be needed. It is that training can no longer be a side effect of tedious work. Until now, nobody had to design it: you assigned simple tasks and waited. When a system does the simple tasks, learning has to be built deliberately, and that changes three things. 1. **A new hire's first year needs rethinking.** Instead of months of mechanical work: mentoring, reviewing what the system produces, and making mistakes in settings where it is safe to do so. 2. **Supervision becomes a skill, not a bureaucratic step.** Checking a system's output requires knowing what a wrong answer looks like. You only learn that by seeing plenty of them. 3. **Implicit knowledge needs to become explicit.** The rules experienced people apply without being able to name them need to be written down: so newcomers can learn them, so they can be checked when a system uses them, and because otherwise they leave the business with whoever retires. This is the slowest part of any serious automation project, and it is useful even if nothing ends up being automated. ## The question to ask When automating a process, the usual question is how much time it saves. It is worth adding a second: what learning disappears along with that work, and where will we rebuild it? It is not an ethical question; it is a continuity question. In ten years, the people who can spot an error in a quotation today will be elsewhere, and someone will need to have learned how to do it. A company that automates without asking saves time now and discovers the problem five years later, when nobody is ready to replace those retiring. ## Sources and further reading **Sources:** Stanford Digital Economy Lab, “Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence”, August 2026 update (ADP payroll data through June 2026) · Economic Innovation Group, “Looking for the Ladder”, an alternative reading based on the interest-rate cycle · Fortune, interview with Erik Brynjolfsson, June 2026 - [Stanford Digital Economy Lab · Canaries in the Coal Mine? (agosto 2026)](https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/) — Updated 12 August 2026. Accessed: 2026-09-18. - [Economic Innovation Group · Looking for the Ladder](https://agglomerations.eig.org/p/looking-for-the-ladder) — Published 14 January 2026. Accessed: 2026-09-18.