An IMF working paper examines how labour demand is changing as generative AI improves. The official publisher record identifies the paper as Working Paper 2026/220, dated October 18. The captured abstract supports a focused account of its findings; the full PDF was not available for this review.
What the researchers measured
The authors analyse more than 700 million online job postings across 110 economies. The abstract reports a decline of nearly 10% in demand for more AI-exposed occupations, while noting that this impact stabilised even as AI capabilities improved.
The evidence concerns job postings. It should not be converted into a count of people who lost jobs or a claim that an equivalent share of employment has disappeared. The distinction between advertised demand and the number of people employed is essential to interpreting the result.
A second signal in AI skills
The researchers also use references to emerging AI skills in postings as a measure of adoption. They report rapid growth in demand for those skills, particularly in more exposed occupations. Firms with stronger demand for AI skills show faster growth in postings in the abstract's account.
Those findings point to changes in the kinds of work being advertised, alongside weaker demand in some exposed occupations. They do not, by themselves, demonstrate that adding an AI keyword causes an employer to expand hiring.
How to read the limits
The publisher labels the paper as research in progress, with views belonging to the authors. A working-paper result is evidence to examine, rather than an institutional forecast or a guaranteed outcome for a particular occupation.
For readers evaluating their own work, the useful question is which tasks and skills employers are actually requesting. Keep the dataset, measured outcome and observational scope attached to any claim about AI and jobs.