Will AI replace jobs? Tasks, jobs and what to watch

01Tasks and jobs are not the same thing

A job is a bundle of tasks, and AI tools take on some tasks faster than others. When a tool handles a task, one of three things can follow: the person spends the time on other tasks, the employer needs fewer people, or the work gets cheaper and demand rises.

The US Bureau of Labor Statistics (BLS) worked through legal jobs as an example in a 2025 case-study article. It expected generative AI to speed up document-related tasks, with paralegals the legal occupation most affected. Yet it still projected lawyers' employment to grow 5.2 percent over 2023 to 2033, reasoning that modest productivity gains would meet strong overall demand for legal services. BLS made these projections from information available in June 2024.

02What has changed so far

The measured effects are small or disputed, and they differ by group.

• The whole labour market: the Yale Budget Lab said on 14 October 2026 that it sees no clear AI-driven shift in the mix of jobs.
• Young workers: a Stanford payroll study, revised on 12 August 2026, finds employment of ages 22 to 25 in AI-exposed occupations is 19 percent below their less-exposed peers. It traces this mainly to less hiring, not more separations, and calls the result descriptive, not causal.
• Recent graduates: the New York Fed reports an unemployment rate of about 5.6 percent for the second quarter of 2026. That page does not attribute it to AI.
• Office work: BLS says generative AI and automation may limit demand in office and administrative support, which it projects to lose 752,100 jobs from 2025 to 2035.

03Three ways AI can show up in a job

The sources describe different patterns, and one job can show more than one.

• Tasks shift. The International Labour Organization (ILO) expects most jobs to be transformed rather than made redundant, because human input is still needed.
• Hiring slows. This is the pattern Stanford reports for the youngest workers in exposed jobs.
• The occupation shrinks. In the BLS projections table, data entry keyers are projected down about 25 percent by 2035. The table does not say how much of each decline is due to AI.

04The case for worry and the case for calm

The case for worry rests on where the numbers already point. Hiring of the youngest workers in exposed jobs has weakened in one payroll dataset, and BLS expects office support work to shrink. If employers keep hiring fewer beginners, the pipeline into many careers narrows even when total jobs hold up.

The case for calm rests on what has not shown up. Yale's first report on the topic, from October 2025, notes that earlier technologies such as computers took years to reshape office work, so a short window may miss the full effect. That cuts both ways: a quiet few years does not prove calm later, and a loud few years does not prove disaster.

05What to watch next

These public sources update on a known schedule, so you can check them yourself.

• BLS projections and the AI exposure table: updated every year, with the 2026 to 2036 set planned for 2027.
• The Yale Budget Lab tracker: it says it will keep updating its analysis.
• The Stanford payroll study: it has been revised since its first version in 2025.
• Your own occupation: find it in the BLS table and read its exposure group and projected change. BLS updates its projections each year.
• The New York Fed graduate page: unemployment and underemployment are updated quarterly, generally in February, May, August and November.