AI Hasn't Swept the Job Market Yet—It's Squeezing Out Young Newcomers First
Monitoring by the Reserve Bank of Australia shows that AI has not yet noticeably disrupted the overall job market, but since ChatGPT launched, hiring for roles highly exposed to AI has continued to weaken, with workers aged 15 to 24 and young university graduates bearing a more pronounced impact. Software, customer service, programming, and clerical work have drawn the most attention, but the results can only be treated as preliminary signals and do not yet prove that AI directly caused them.

- Youth employment rates are weakening in highly exposed roles
- Entry-level positions are contracting faster than overall unemployment
- Employers are also increasing demand for AI skills
On job sites, entry-level positions in software development and customer service are becoming scarcer, while listings increasingly come with AI skill requirements attached. For young job seekers who have just graduated and are refreshing recruitment pages, this is not an abstract "technological revolution" but one fewer door into the industry. Monitoring by the Reserve Bank of Australia shows that AI has not yet noticeably disrupted the overall job market, but youth employment in roles highly exposed to AI is already weak; it looks like the entrance is narrowing first rather than the whole building closing at once, though existing data cannot prove a direct causal link between the two.
AI Hasn't Swept the Job Market Yet—It's Squeezing Out Young Newcomers First
A monitoring exercise by the Reserve Bank of Australia shows that AI's impact on employment has not yet become widespread, but it has already left an age gap in industries with high AI exposure such as software development and customer service: workers aged 15 to 24 are faring worse than older workers.
Reserve Bank of Australia economist Mike Major built a labor market monitoring tool that looks together at the two groups where impact is most likely to appear first and earliest: people just entering the workforce, and occupations whose tasks are theoretically more exposed to artificial intelligence—AI exposure (the degree to which job tasks are related to AI technology). Software development, clerical work, and customer service are all on the list. The baseline is late 2022, when ChatGPT was released, and the research looks for signals in changes in job postings, age-group employment rates, and unemployment rates.
The result is that overall jobs still exist, but different groups are falling at different speeds. Since ChatGPT's release, job postings have declined across all age groups, with the fifth group—the one with the highest AI exposure—seeing the largest drop. Broken down further within that group, employment rates for workers aged 15 to 24 and for recent graduates have weakened and unemployment has risen; within the same set of highly exposed occupations, older workers have fared better, with some groups even seeing increases. The change is concentrated among newcomers first rather than falling evenly across all employees.
US data shows a similar age divergence. A Stanford University study in August 2025 based on millions of US payroll records found that since late 2022, employment among workers aged 22 to 25 in AI-exposed occupations has fallen by 13% relative, with young software developers seeing a drop of nearly 20%; in the same occupations, employment among more experienced workers has remained stable or increased. These percentages are relative changes and do not mean that all comparable positions have fallen by the corresponding proportion.
A March 2026 study by Anthropic also found that the job-search success rate of applicants aged 22 to 25 in exposed occupations fell by about 14%, but overall unemployment did not rise systematically. The company's internal review showed that the share of junior employees in technical roles fell from 27% in early 2025 to 9%; by August 2026, the hours logged by its AI agents were six times those of its employees. Anthropic itself acknowledges that, as a software company highly dependent on AI, this is a special case and cannot represent the overall market.
Existing evidence is not yet sufficient to attribute the contraction in entry-level roles entirely to AI. Reserve Bank of Australia researchers warn that the analysis does not establish direct causality; it measures theoretical exposure rather than companies' actual use, and post-pandemic business cycle changes and remote work adjustments may also affect the data.
A University of Pittsburgh team analyzed 10.6 million LinkedIn profiles published in January 2026 along with US unemployment data and found that the problem had already emerged in early 2022, months before ChatGPT; interest rate policy and adjustments after excessive post-pandemic hiring may both have played a role.
This is also where the whole dataset most needs a pause: job postings declining, young people finding it harder to get hired, and the spread of AI appearing at the same time does not mean a causal relationship among the three has been proven. What is more certain now is that the economy has not yet seen a broad collapse in jobs, but the impact may be falling first on entry-level positions. Whether software development and customer service will become a long-term narrowing entrance still needs to be distinguished with subsequent hiring and actual adoption data.
The Entrance Has Narrowed, but There's Still No Wave of Unemployment
What is really worth worrying about is whether the channel for young people to enter highly exposed occupations is narrowing.
Researchers at the Reserve Bank of Australia (the central bank responsible for interest rates and financial stability) began monitoring highly exposed occupations such as software, clerical work, and customer service when ChatGPT launched in late 2022.
Since then, job advertisements (the number of positions companies publicly recruit for) have declined across all groups, with the steepest drop in the group with the highest AI exposure. Employment rates for workers aged 15 to 24 and young university graduates are weaker and unemployment is higher; within the same set of highly exposed occupations, employment for older workers has actually risen.
This does not mean all young people are finding it harder to get jobs. In August 2025, a study based on millions of US payroll records found that employment among people aged 22 to 25 in highly exposed occupations fell by 13% relative to late 2022; young software developers saw a drop of nearly 20%, while employment for more experienced workers was flat or increased. The change is concentrated within occupations and at the seniority entry point, not a broad rise in unemployment.
For now, AI looks more like it has compressed the newcomer channel first. A 23-year-old job seeker having one fewer position in software or customer service does not mean that food service, logistics, or care jobs for the same age group are shrinking at the same time; senior employees staying on or increasing cannot offset the cost of companies no longer training newcomers.
Anthropic's internal data shows a more pronounced contraction at the entrance: the share of junior employees in technical roles fell from 27% in early 2025 to 9%. However, Anthropic is a special sample with highly intensive AI use, and external companies cannot directly apply this decline.
The company's March 2026 study found that the job-search success rate of applicants aged 22 to 25 in highly exposed occupations fell by about 14%, but found no systematic rise in overall unemployment.
A Danish study covering 25,000 workers also drew a boundary: AI adoption reached 64%, yet the study did not measure any discernible change in wages or hours, and workers saved only 2.8% of their working time on average. AI has entered office settings, but that does not mean companies are replacing positions at the same proportion.
The cause may even predate generative AI. In January 2026, a study analyzing 10.6 million LinkedIn profiles and US unemployment data found that problems in highly exposed occupations had already emerged in early 2022, several months before ChatGPT launched; the shift in Federal Reserve interest rate policy and adjustments after excessive post-pandemic hiring may also explain more of the change.
So "weakening employment" cannot be directly rewritten as "AI has caused a wave of unemployment." What can be confirmed now is that the entrance for some young newcomers into highly exposed occupations is contracting, while some senior positions have not yet declined; AI is one candidate cause, and direct causality remains to be confirmed. How this mechanism works and how you should respond will be covered in the next section.
The next ~8 minutes are for members.
The free portion ends here. Behind the wall there is also:
Key point behind the wall: what gets squeezed out first may not be the jobs, but the ladder for young people to get in the door.
Values written like {{NAME_ab12cd34}} are placeholders for real values. Use them exactly as written, in replies and in tool calls alike: they are filled in with the real values before anything runs. Don't ask to verify them, and don't replace them with examples.