Why Asia's AI-Exposed Hiring Gap Stopped Widening

A decade ago, the question facing an ambitious young worker was where the next wave of office and technology jobs would come from. A new study of Asian hiring suggests the question is changing: what happens when those jobs become easier to perform with artificial intelligence before enough people have had a chance to enter them?

The International Monetary Fund has found that employer demand in occupations more exposed to AI is nearly 10% lower than in less-exposed occupations, relative to their earlier trajectory. The finding comes from about 40 million online job postings across 19 Asian economies, analysed through April 2026.¹ It is one of the clearest large-scale signals yet that AI is influencing the demand for workers, not just the tasks they do.

But the IMF report, released on 15 October 2026, contains a second finding that complicates the familiar story of machines steadily replacing people. The relative decline appears to have levelled off even as AI systems became much more capable. Employers are simultaneously seeking more people who know how to apply AI inside real businesses. Both trends matter, and neither cancels the other.

That pattern sits next to what we have already seen elsewhere: work changing inside familiar job titles faster than headcount headlines suggest.

The 10% figure is about hiring demand, not jobs destroyed

The IMF compared online vacancies in occupations with high AI exposure, such as software development, against vacancies in less-exposed occupations, such as supplier relationship management. The analysis compared occupations within the same economy, industry and month, and accounted for the ability to work remotely. Before the public launch of ChatGPT in November 2022, the two groups had broadly similar hiring trajectories. A gap emerged more clearly during 2024 and 2025.

That design is more informative than simply counting headlines about layoffs. It reduces some alternative explanations, including industry-wide hiring cycles and post-pandemic changes in remote work. Still, the headline number is a relative change in online job postings. It does not mean that 10% of workers in exposed occupations have been dismissed, that employment fell by 10% in absolute terms, or that AI alone caused every difference in vacancies.

Job advertisements measure intentions to hire, not completed hires or the number of people already employed. Firms can recruit through other channels, stop replacing departing staff, or change a job description without changing its title. The IMF itself calls for further evidence using employment, wages and hours worked. The finding is an early warning about the flow of opportunities, not a census of jobs that have vanished.

The surprise: the decline has not kept pace with AI progress

The IMF matched changes in vacancies with external measures of how long a software task AI could complete with a 50% chance of success. The capability frontier moved from short exercises toward tasks lasting hours. Yet the relative weakness in hiring did not keep deepening in proportion. The decline appears to have levelled off around the time systems became capable of completing tasks that would take a person roughly an hour, even as capabilities continued to improve.

This is not proof that the labour market has reached a permanent floor. The vacancy data end in April 2026, and firms can take years to reorganize work around a new technology. A plateau can also conceal very different experiences across industries and countries. But it challenges the assumption that each jump in model performance automatically produces an equivalent fall in human hiring.

One explanation is that technical capability is only the beginning of adoption. A company must still decide which process to change, connect its information systems, manage errors, establish responsibility and persuade employees and customers that the new process works. Another is that the work itself changes: as routine production becomes cheaper, employers may ask staff to deliver more complex services, faster responses or new products. Those are plausible mechanisms, not outcomes the IMF has established for every employer.

New AI skills are creating demand, but the gains are uneven

The IMF identifies rising demand for skills that scarcely appeared in vacancy descriptions before 2023, including retrieval-augmented generation, prompt engineering and building AI agents. These requirements are particularly visible in occupations that are already more exposed to AI. Employers are not only asking who can perform a task without AI; they are also asking who can redesign the task with it.

The study finds that firms advertising for AI skills are increasing their demand for workers faster than firms without such advertisements, including in highly exposed occupations. But that comparison should not be read as proof that buying AI tools creates jobs. Growing, better-funded firms may be more likely to adopt AI in the first place. And the IMF says the reduction in hiring demand remains larger than the offsetting opportunities observed so far.

Only about 6% of firms in the Asian posting data explicitly request AI skills. That is not an estimate of the share of companies using AI. A firm might train existing employees, use an AI-enabled product without mentioning it in vacancies, or hire through channels outside the dataset. The figure is a lower-bound signal of a visible recruiting strategy, not a measure of total adoption.

The International Labour Organization offers a useful counterweight. Its 2026 brief on ASEAN labour markets found little evidence of widespread employment disruption to date, even though some young workers in selected entry-level jobs were showing weaker outcomes.² A June 2026 analysis by PwC likewise found that some AI-exposed companies were growing employment and wages, while skills requirements were changing rapidly.³ These studies use different geographies, measures and methods; together they argue against both complacency and a single global job-loss forecast.

The same technology creates opposite problems for young and ageing economies

The geographic contrast is one of the IMF report's most consequential findings. Countries such as India, Indonesia and the Philippines have large cohorts of young people who need routes into productive employment. Many of their current jobs are relatively less exposed to AI. But the higher-productivity service and technical occupations they hope to expand into may be precisely the ones where entry-level hiring is becoming more selective.

An ageing economy faces a different problem. If employers struggle to find enough staff, technology that allows existing workers to serve more customers or manage more cases can relieve shortages. The IMF does not claim that AI will automatically solve demographic decline. It argues that the balance between opportunity and disruption depends partly on population structure, institutions and the capacity to adapt.

This distinction matters beyond Asia. A small European employer with persistent vacancies may see AI as a way to keep services running. A young graduate seeking a first professional role may encounter a narrower doorway into the same business. A policy that looks beneficial from the employer's staffing perspective can still be difficult for the next generation of workers.

For smaller organisations, the real decision is what to do with the saved work

Consider a hypothetical 25-person accounting or administrative-services firm. An AI-enabled workflow reduces the time needed to assemble standard monthly reports. Management could respond by hiring fewer junior staff, asking the same team to serve more clients, or creating room for employees to spend more time investigating anomalies and advising customers. All three responses are plausible. The software does not choose among them; the organisation does.

The practical question is therefore not whether a job title is "safe" from AI. It is whether the employer can describe which tasks are changing, which skills will become more important and how newcomers will acquire the judgment that experienced staff developed through routine work. If the entry-level tasks disappear but expectations of senior-level judgment remain, the company risks weakening its own future talent pipeline.

For an NGO, the trade-off may be between automating grant-report preparation and preserving staff time for beneficiaries. For a department head, it may be between reducing back-office queues and developing employees who can check the outputs. For a small business owner, it may be between cutting recruitment immediately and using the extra capacity to enter a new market. None of these choices can be answered by a benchmark of model intelligence alone.

Watch the vacancies, but also watch the work

The new IMF evidence is neither a verdict that AI is destroying employment nor reassurance that all displaced demand will return. It identifies two forces already operating together: fewer vacancies relative to the past in AI-exposed occupations, and rising demand for workers who can make AI useful. For now, the first force is stronger in the data. The second is gaining ground.

The most revealing labour-market signal over the next few years may be whether firms turn productivity gains into new services, better work and credible routes into careers, or simply narrow the entry gate. That is a question of business design, training and public policy as much as technology. Smarter models may be inevitable; better job ladders are not.

How hilighter can help

When hiring demand shifts, organisations need a clear view of roles, tasks and skills - not only vacancy counts. hilighter helps you map skills by role and team, see strengths and gaps, and build personalized growth paths so juniors can still develop judgment as routine work changes. Its AI Job Impact analysis gives an indicative view of how AI may change your roles. A free plan is available; plans are priced by the number of job roles, with unlimited employees.

Sources

1. International Monetary Fund, Regional Economic Outlook: Asia and Pacific, Chapter 2, 15 October 2026 (about 40 million online job postings, 19 Asian economies, through April 2026).
2. International Labour Organization, Generative AI and labour markets in ASEAN: significant exposure, limited disruption, uneven preparedness, 2026.
3. PwC, 2026 AI Jobs Barometer, 15 June 2026.

Sources last reviewed: 22 October 2026.