Staff at Australia's central bank have started tracking where AI would first show up in the job market, and the early answer is among young workers in the most exposed jobs. Those are software developers, clerical and administrative workers and call-centre staff. Job ads in those occupations have fallen more than in other jobs since ChatGPT launched in November 2022. Employment for 15 to 24 year olds in them is now below its level then, while employment for workers aged 25 to 64 has risen. For the economy as a whole, the author finds little disruption so far and calls these signs tentative.
Why the young are the first to feel it
The same analysis finds that people in exposed jobs are not losing them faster, and the unemployed who used to work in those jobs are not finding new work any more slowly. So AI is not pushing people out. It is shrinking the number of beginner positions that open up.
My reading is that this is how a change like this usually starts. A manager who has a task AI can now do does not fire anyone. She waits for the next person to leave and does not refill the seat, or she hires one junior where she would have hired two. Beginners are the ones who fill those seats, because routine, clearly defined work is what beginners are given first, and it is also the work AI does best. Experienced people are protected by the part of the job that is judgment, and that is why total employment looks fine while the entrance narrows.
What the tellers show
Bank tellers saw the same thing before. When cash machines spread through American banks, each branch needed fewer tellers, yet teller jobs held up because cheaper branches meant more branches. The half of that story that matters for beginners is what happened to the tellers who stayed: they were doing different work. Cash handling mattered less, and banks wanted tellers who could build relationships with customers, especially small businesses, and sell them more services. The beginner job survived because it changed into something a machine could not do.
For example, a small online shop answers 300 customer emails a day with three junior staff. If an AI drafts each reply from the order history and a person checks it, I would expect one person to get through roughly twice as many emails. The owner now has two choices. She can hire one junior instead of three and stop there. Or she can keep all three and use the time to phone customers whose orders went wrong, handle returns before people ask, and answer in minutes instead of hours. The second owner has the same three jobs, but they are different jobs from the ones she would have hired for two years ago.
How much of this is AI
Not all of it. Young Australians' unemployment rate across all jobs rose from about 7% in 2022 to around 11% in early 2026, and 2022 was an unusually tight year for hiring, so part of the fall is a cooling economy. A separate US analysis found that trouble in AI-exposed jobs began in early 2022, before ChatGPT existed. But a Stanford team found that even after removing company-level shocks, workers aged 22 to 25 in the most exposed jobs had a 13% relative decline in employment. Together they point to AI as a real part of the story and not the whole of it. We reported the same pattern in software in September, when entry-level software jobs had fallen 20% since ChatGPT launched.
The deeper risk is that beginners used to learn the job by doing the routine work. If AI takes that work and nothing replaces it, firms get a shortage of experienced people in five years that they cannot hire their way out of.
If you manage a team, decide now what a beginner on it will do all day. The routine part is going to AI, and what remains is dealing with customers and checking the work, which is what your next hire will be paid for.