IMF Finds AI Is Changing Asia's Hiring as Programmer Demand Falls About 10%

Hiring pressure rises where AI can do more tasks

Employer demand for programmers fell about 10% relative to less exposed occupations during 2024 and 2025, according to International Monetary Fund research presented in Bangkok on October 12, 2026. The analysis, described as covering 700 million job advertisements, also found that demand for artificial intelligence skills was roughly twice as high in occupations where AI could assist with more tasks.

The findings point to a change in recruitment rather than evidence that one in 10 programmers has lost a job. Divya Kirti, a senior economist in the IMF's Asia and Pacific Department, said employers still wanted programmers, particularly people able to use AI to accomplish more. Hiring demand in more exposed occupations had not continued to fall sharply as the technology improved.

The IMF's official session listing scheduled the presentation, Labor Markets in Asia: Challenges and Opportunities in the Age of AI, for 2:30 p.m. to 2:55 p.m. in room G-03. It placed the session under the Asia and Pacific Department and described research drawing on millions of online job postings across Asia.

The central finding is narrower, and more useful, than a prediction of mass unemployment: employers are changing which skills they seek, while some occupations face weaker recruitment. Aggregate hiring had not deteriorated in the account presented, and the research had not established clear effects on wages or inflation.

Those distinctions matter for governments deciding whether to focus on supporting displaced workers, preparing new entrants or helping existing employees adapt. A decline in vacancies can restrict career opportunities even before it produces a visible fall in the number of people already employed.

What employers mean by AI skills

Kirti identified three groups of capabilities sought by employers. The first is general AI literacy, meaning an understanding of the technology and an ability to apply it to everyday work. The second covers specialised capabilities, including prompt engineering and AI agents. The third involves connecting AI with a company's internal knowledge and data.

Prompt engineering means designing instructions that help an AI system produce the intended result. AI agents can pursue objectives through sequences of tasks. Integrating company knowledge goes further than using a public chatbot: it connects the technology with information and expertise a business has accumulated.

The roughly twofold difference in AI skill demand compares more exposed occupations with less exposed ones. It does not mean demand doubled over time. The same occupations facing greater pressure on hiring were also those where employers placed greater value on AI capabilities.

A separate IMF staff discussion note, Bridging Skill Gaps, helps explain why this does not necessarily produce a surge in recruitment of AI specialists. Many businesses can adopt ready made tools or purchase services instead of building large technical teams.

The note distinguishes people who use AI applications from those who build and deploy models. In the United States in 2024, roughly half of advertisements requesting AI skills mentioned only user skills, one quarter mentioned only developer skills, and one quarter sought both. That distribution suggests education needs to reach beyond computing departments into occupations where AI is becoming a routine work tool.

The change became clearer after ChatGPT

The Asian recruitment analysis found little initial change following ChatGPT's launch in late 2022. The relative decline in programmer hiring became clearer during 2024 and 2025. That timing qualifies the original account's broader description of changes following the arrival of large language models.

Kirti described a progression from systems handling tasks that would take a person seconds to tools capable of debugging code that might take a programmer an hour. More recent capabilities included assignments comparable to a day's work, such as building models to analyse business data and identify factors affecting sales.

The evidence spans several distinct periods:

• Before 2015: AI skills appeared in fewer than 1% of US job postings in the IMF discussion note's series.
• Late 2022: ChatGPT launched, providing a reference point for studies of generative AI and employment.
• 2024 and 2025: Programmer recruitment demand fell about 10% relative to less exposed occupations in the Asian analysis.
• By 2025: Almost 5% of US postings requested AI skills in the discussion note.
• October 12, 2026: The IMF presented the Asian findings at the annual meetings in Bangkok.

The US posting figures indicate an expansion from fewer than one vacancy in 100 to almost one in 20. They measure the share of advertisements mentioning AI skills, not the share of workers using AI or the number of jobs the technology has created.

Why stronger skill demand does not guarantee more jobs

The IMF discussion note draws a distinction between new skills generally and AI skills specifically. About one in 10 vacancies in advanced economies requires at least one new skill, compared with roughly one in 20 in emerging market economies. These requirements are concentrated in professional, technical and managerial occupations.

New skills generally were associated with higher wages and stronger local employment. Benefits were largest for workers with high and low skill levels, with no significant gains for those in the middle. One mechanism identified in the note is that increased spending by better paid workers can support demand for services provided by workers with lower skill levels.

AI skills presented a different pattern. Their spread had not increased aggregate employment in US local labor markets. For occupations highly exposed to AI, with limited opportunities for the technology to assist rather than substitute for workers, employment was estimated to be 3.6% lower after five years in regions with greater AI skill demand.

That estimate reflects an observed increase of 0.57 percentage point in relevant skill postings. The model associated a one percentage point increase with employment levels 6.3% below those in comparison regions. These are findings about differences between regions and occupation groups, not a forecast that Asia will lose 3.6% of its jobs.

The key concept is whether AI complements a worker. A job can be highly exposed because many tasks are accessible to AI, yet still benefit if the technology helps a person perform those tasks. The note found no significant employment effects for highly exposed occupations with greater scope for that kind of assistance.

Young workers face an uneven picture

The Asian study did not find clear evidence that younger or entry level workers were more severely affected than experienced employees. That does not settle the debate about graduate opportunities. Other research discussed by the IMF points to a different experience in the United States.

The discussion note cites a US study finding a 13% relative decline in employment among workers aged 22 to 25 in the most AI exposed occupations since ChatGPT's release. Employment remained stable or grew for more experienced workers in the same occupations and for workers in less exposed fields.

The findings concern different populations and datasets. The absence of a clear age effect in the Asian analysis should not be read as disproving the US result, nor should the US result be presented as a measured decline across Asia.

Singapore President Tharman Shanmugaratnam, who co chairs the World Bank's Jobs Council, warned that declining graduate recruitment could weaken the first step into a career. He described the concern at the meetings:

There is a particular problem that we have to address very soon ... and that is that, the first rung in employment, your first job upon graduation, is now seeing less demand.

He called for employment support used during the pandemic and financial crises to become a permanent policy tool, including incentives for businesses to hire graduates. His argument was that displacement would occur task by task, leaving time for learning at work and more applied university training.

Asia faces both workforce decline and a need for new jobs

For developing Asia, Kirti put the employment challenge at 150 million new jobs over the next decade as large young populations enter the labor market. Spread evenly, that would mean an average of 15 million additional jobs a year. The challenge is to create opportunities in more productive sectors even as AI changes recruitment in those sectors.

Wealthier Asian economies face a different problem. Findings presented in an IMF book indicate that China, Japan and South Korea could lose roughly a quarter of their working age populations by 2050. AI could help compensate for fewer workers, though it would not necessarily resolve every labor shortage.

IMF Managing Director Kristalina Georgieva estimated that AI could lift growth by 0.5 to one percentage point in advanced Asian economies and by 0.2 to 0.8 percentage point in emerging Asian economies. She said the gains could more than offset demographic pressure if countries captured the productivity benefits.

IMF Asia and Pacific Department Director Krishna Srinivasan described the policy choice as whether the technology assists workers or replaces jobs:

What matters is whether AI augments workers or automates jobs. Advanced economies, for sure, are more exposed, but they are also better placed to benefit. Emerging markets are less exposed, but also less in a position to capture the upside.

Srinivasan said reforms improving skills and productivity could bring adoption forward by a decade or more. Lower immediate exposure in developing economies, partly reflecting employment in agriculture, therefore does not guarantee a stronger position over time.

Exposure estimates need careful reading

The IMF discussion note estimates that 40% of global employment is potentially exposed to AI. Its figures are 60% for advanced economies, 40% for emerging markets and 28% for countries with low incomes.

Accounts of Georgieva's remarks instead put exposure in countries with low incomes at 23%, while giving the same 60% estimate for advanced economies. The five percentage point difference is not reconciled in the reported remarks. Neither figure should be treated as a projected rate of job losses.

Georgieva described the global 40% share as jobs that could eventually be augmented, downgraded or eliminated. That category includes work made more productive as well as work reduced or removed.

A separate discrepancy concerns current US AI requirements. Accounts of her remarks cited a study finding that 10% of US jobs already required AI skills. The official discussion note records almost 5% of US job postings mentioning AI skills by 2025, while its one in 10 figure concerns any new skill across advanced economies. Those measures cannot be treated as interchangeable without further detail on definitions and periods.

Wage premiums are not the same as wage growth

Kirti said the Asian research had not yet found clear evidence of AI effects on wage levels. It also did not establish whether AI would raise or lower inflation. Those unanswered questions limit claims that weaker hiring has already reduced labor costs or consumer prices.

The discussion note nevertheless found that advertisements requesting AI skills often offered higher pay than other vacancies within the same occupation. In the United Kingdom, both developer and user skills were associated with posted wage premiums of about 7.5% to 8%. In the United States, developer skills carried premiums above 8%, while user skills carried a smaller premium near 2%.

These comparisons concern advertised pay for different vacancies. They do not show that AI adoption increased the wages of all existing workers. The note's analysis of AI skill diffusion found no significant wage effect in general, although workers in highly exposed occupations with little scope for assistance experienced modestly lower wages two to three years after the skills appeared.

Daniel Kostzer, chief economist of the International Trade Union Confederation, raised another dimension at the session. Observations from labor organisations pointed to displacement risks in some occupations and greater workloads or work intensity in others. He asked how workers could participate in adoption and adjustment.

That distinction broadens the policy issue beyond headcounts. A business can retain employees while changing their responsibilities, pace of work or bargaining position. Kirti also identified potential limits to adoption, including high energy costs and the possibility that AI services were being priced below their full costs.

Growth gains bring questions about who benefits

Georgieva said AI investment and trade were helping keep global growth around 3% despite an energy supply shock. Asian suppliers of hardware and infrastructure, including businesses in Thailand, were benefiting from the investment boom.

Stronger output does not guarantee stronger employment in the same occupations. Shanmugaratnam warned that job creation was weak even in growing economies such as China, India and developing Southeast Asia. Georgieva also described expanding lower paid work in construction and data centres alongside shrinking routine tasks.

World Bank President Ajay Banga pointed to practical applications he called "small AI" for farmers, healthcare systems and educators in emerging markets. These applications need not depend on large language models, and the bank's focus was on productivity gains in sectors where many people already work.

Banga said the private sector generates 75% to 85% of jobs and would need to absorb much of the coming increase in young workers. Reports differ sharply on his demographic estimate: one gives 1.2 billion people turning 18 over the next 15 years, while another gives 1.2 million. The figure cannot be reconciled from those accounts.

Georgieva also warned about public finances if economic activity shifts from labor toward capital. Governments could face pressure on tax revenue at the same time workers need more support. No quantified revenue loss or specific tax reform was announced in the accounts of her remarks.

What governments can do, and what comes next

The IMF discussion note recommends accessible retraining, stronger unemployment insurance and support for movement between occupations and regions. Affordable housing, remote work and active employment policies could help workers reach opportunities rather than remain tied to declining demand.

Its education recommendations include cognitive, creative and technical skills that help people work with AI, alongside better career guidance, apprenticeships and internships. Kirti's presentation similarly called for education linked to practical work experience and support for occupational transitions.

The note also recommends cooperation between employers and unions, limits on agreements that restrict employees from joining competitors, and scrutiny of business concentration that could slow the spread of skills. These proposals address access to opportunities as well as the content of training.

The broader Asian growth discussion extended beyond AI. Georgieva said deeper regional trade agreements and fewer barriers other than tariffs could raise Asia's gross domestic product by about 1.8% over the long term. That estimate concerns the level of economic output, unlike the percentage point estimates for AI's contribution to growth.

The next scheduled release identified in the meeting coverage was the IMF's updated economic outlook on October 13, including forecasts for South Korea and other countries. The employment findings did not set a deadline for new support measures or determine how many workers would need to change occupations.

For now, the strongest conclusion is that recruitment is already changing, while the scale of lasting displacement remains uncertain. The IMF's recommendations seek to preserve routes into work and help people adapt before weaker vacancy demand becomes a longer period without employment.

Key Points

• Programmer hiring demand fell about 10% relative to less exposed occupations during 2024 and 2025, not a measured 10% layoff rate.
• Demand for AI skills was roughly twice as high in more exposed occupations as in less exposed ones.
• The Asian analysis found no clear evidence yet of wage effects or a greater impact on younger workers.
• Separate US research cited by the IMF found employment pressure in occupations where AI has limited scope to assist workers.
• Developing Asia needs 150 million new jobs over a decade, while China, Japan and South Korea face large workforce declines by 2050.
• The IMF recommends practical education, accessible retraining, worker mobility and stronger social protection.