AI and jobs in Asia: what the fall in job adverts means

The design narrows alternatives but cannot prove causation

The comparison holds constant shocks shared within the same economy, industry and month. It also uses economy–industry–occupation fixed effects, adjusts for pre-launch remote-work exposure over time and clusters uncertainty by occupation. More-exposed occupations are defined with an existing complementarity-adjusted exposure index, rather than by whether a particular employer says it replaced a worker with AI.

The sample was selected using Lightcast coverage from May 2025 to April 2026. Economies needed at least 1,000 postings and either active worker profiles covering 5% of the labour force or more than 25,000 postings. Staffing-agency adverts and postings without an identifiable employer industry were excluded. These rules improve consistency, but online recruitment coverage varies by country and can under-represent informal work and hiring through other channels.[1][2]

The skills signal is promising, not a guarantee for workers

The IMF also identified 51 emerging AI skills, including prompt engineering, retrieval-augmented generation and agentic AI. About 6% of firms in the Asian sample asked for an AI skill. Among firms already advertising those skills, demand grew more strongly for highly exposed occupations than among non-adopting firms. This could help explain why the overall relative decline stopped deepening.

For graduates and workers, the practical message is not simply ‘learn prompting’. Employers may be redesigning roles while hiring fewer people into some entry routes, and the report does not track whether training leads to durable employment or higher pay. Better evidence would link postings to hires, wages, hours and worker histories, test results outside online-recruitment-heavy sectors, and examine whether the gap persists through a downturn.

A separate descriptive analysis of 65 million online career histories found that only about 5% of profiles moved across occupations in a year, with no recent shift towards less AI-exposed work. That is a warning about adjustment speed, not proof that people were trapped by AI: profile coverage is selective, simultaneous jobs were excluded from the denominator and the data do not reveal why a person moved, stayed or left employment.[1][2]

How to use the finding without turning it into a personal forecast

The same regional result can raise different questions for different readers. Someone entering work needs to know whether suitable entry roles are being advertised and what skills they require. An employer needs to distinguish a change in recruitment from a change in the work done by existing staff. The IMF’s regional comparison cannot answer either question for a particular person or firm. Its discussion of younger and ageing economies also underlines why a single Asian average cannot describe every local labour market.

Our interpretation is to treat the report as a reason to examine the next layer of evidence. Look for country and occupation breakdowns, coverage of informal recruitment, actual hiring and whether new skill requirements are accompanied by training. These are follow-up questions, not benefits or harms demonstrated by the posting figures. A useful assessment would separate adverts, hires and the tasks those people perform. Combining them into one job-loss estimate would hide the distinctions the report makes important.[1][2]