In July 2026, the International Labour Organization publishes an unprecedented assessment of Southeast Asian labour markets in the face of generative AI. The jobs exposed are numerous, mass layoffs remain rare, but countries’ capacity to absorb what comes next is profoundly unequal. The outcome depends on institutions that decide how gains are distributed, more than on the technology itself.

The essentials

  • The ILO concludes in July 2026 that there is little evidence of widespread employment disruption linked to generative AI in ASEAN, despite significant exposure of many occupations (ILO, Generative AI and labour markets in ASEAN, 8 July 2026).
  • Exposure to generative AI tools varies sharply by country: Singapore is most prepared; Malaysia, Indonesia, the Philippines and Vietnam have substantial foundations but retain variable gaps.
  • Adoption and its effects depend also on skills, infrastructure, business capacity, governance, social protection and social dialogue.
  • The window is open, and provisional: countries that build these institutions now will set the conditions in which tens of millions of young workers enter the market.

Exposed occupations transform before they disappear

Generative AI can transform or automate cognitive and administrative tasks, but its observed use remains concentrated in technology-intensive and knowledge-intensive jobs. In ASEAN, these tasks are concentrated in financial services, call centres, data entry and certain segments of law and accounting. The ILO estimates that a substantial share of office employment is exposed to significant transformation of its content.

But exposed does not mean eliminated. The history of previous waves of automation confirms this: when a technology takes over tasks, the workers who remain supervise it, correct it, and direct it. This phenomenon, often called augmentation rather than replacement, is already observable in Singaporean and Malaysian companies that have deployed automatic text generation tools in their customer support teams. Headcount has not fallen; low-value-added tasks have declined, verification and escalation tasks have increased.

The problem is that this reshaping requires two conditions that many ASEAN countries meet poorly: workers capable of moving to more complex tasks, and employers incentivized to support them in doing so rather than replace them with external providers.

Singapore and the Philippines read the same technology in incomparable realities

The inequality of preparedness between ASEAN member states is one of the most striking facts in the ILO report. Singapore has one of the most well-funded continuous training systems in Asia, a dense institutional fabric and structured social dialogue. Malaysia has pursued a national digitalization policy since 2023 that explicitly integrates upskilling. At the other end, the Philippines and Indonesia combine strong sectoral exposure—call centres and related activities employed approximately 652,000 people in 2019 according to the Philippine Statistics Authority—with still-fragmented retraining mechanisms.

This asymmetry reflects a difference in institutional capacity accumulated over a long time, not a difference in political will. Gaps in public investment in adult training and retraining remain significant between countries in the region. Generative AI did not create this gap, but it will likely amplify it, as multinational companies choose where to locate the functions they retain versus those they automate.

Vietnam represents an instructive intermediate case. The country has attracted significant investment in recent years in electronics manufacturing and digital services, partly because its wage costs remain low. But part of its competitiveness rests precisely on data processing and back-office tasks that generative AI is beginning to take over. Asia is gaining in volume in world trade, but the comparative advantages that underpin this position are being reconfigured, and the least institutionally prepared countries will bear the cost first.

The region’s demographic advantage can become a liability

ASEAN is among the youngest regions in the world. Indonesia, the Philippines and Vietnam record significant annual flows of new workers entering the labour market. This demographic advantage, often presented as an asset against ageing economies in Europe or Japan, is real, but conditional.

It assumes that available jobs are numerous enough and of sufficient quality to absorb this workforce. Generative AI introduces new pressure precisely on office and service jobs that these young workers targeted as a gateway to the middle class. In previous decades, the classical path ran from agriculture to manufacturing, then from manufacturing to services. If information processing services automate before the economies in question complete this transition, the next step on the path is not yet built.

Southeast Asia shares this tension with other regions. An article on Africa posed an analogous dynamic: economies that have not yet completed their industrial transition can leverage AI to skip steps, but they also risk seeing their comparative advantages erode before building new ones. The answer depends largely on institutional choices made now, before net job losses become visible in the statistics.

The role of institutions as technology arrives

The ILO report does not merely map exposure. It indicates that adaptive capacities, skills, infrastructure and institutions can lead to different national trajectories. Three levers stand out in its analysis.

The ILO cites strengthening social dialogue mechanisms among governance priorities. In economies where unions participate in decisions about introducing new technologies—on pace, conditions and trade-offs—workers can steer adoption toward capacity augmentation rather than replacement. Singapore has institutionalized this dialogue through its National Wages Council, which brings together government, employers and unions on major policy directions. This tripartism frames the conditions for technology adoption without slowing its pace.

The ILO recommends expanding opportunities for upskilling and retraining, as well as employment services. The logic is straightforward: if generative AI allows a company to produce more with the same workforce, some of the gains must finance the upskilling that makes this productivity sustainable. Without a binding mechanism, companies externalize this cost, onto the state, onto workers themselves, or onto the labour market that absorbs the laid-off.

The ILO particularly calls for strengthening regional qualifications and skills frameworks linked to AI, aligned with national systems. When a worker’s retraining is certified in a way recognized across an entire sector, not just by their employer, their capacity to change companies or roles increases. The Philippines launched in 2025 a micro-certification programme in digital skills via their technical education authority (TESDA), targeting call centre workers. Scale remains modest against the size of the sector, but the model exists and can be scaled up.

The two possible trajectories by 2035 and what distinguishes them

The ILO does not project a ten-year numerical scenario, and none of the available sources do so with sufficient credibility. But the architecture of forces in play allows us to distinguish two conditional trajectories, whose signals of arbitration are already partially observable.

In the first scenario, ASEAN countries with solid bargaining institutions frame AI adoption through rules that condition the introduction of new tools to an employer-financed upskilling plan. The technology improves productivity, gains are partly redistributed as higher wages or reduced hours, and the region’s demographic advantage converts into improved job quality. This scenario is conditional on specific policy decisions, made before net job losses become visible.

In the second, countries adopt generative AI tools, often purchased from international suppliers, sometimes subsidized by national digitalization policies, without building in parallel the institutions that organize retraining and negotiation. The unequal exposure documented by the ILO then translates into a lasting widening of the gap between prepared and vulnerable countries. Workers in the most exposed sectors absorb wage pressure without the tools to negotiate it or the training to respond to it.

What distinguishes these two trajectories is less the speed of technological adoption than the state’s capacity to build institutions ahead of effects. The Asian Development Bank Institute (ADBI), in an October 2026 note, identifies a third collective need often underestimated: measurement. Countries that do not know which sectors are transforming, or at what pace wages are moving in exposed occupations relative to others, do not have the information needed to intervene at the right time. Statistical monitoring tools, tracking median wages in occupations heavily exposed to generative AI, the share of workers with access to recognized training in high-exposure sectors, would allow divergence to be detected before it becomes irreversible.

A dedicated ASEAN regional fund for institutional preparation, prioritizing countries whose documented gap is greatest, is one path mentioned by the ILO. The idea is straightforward in principle: pool a capacity that the most vulnerable economies cannot build alone, within the timeframe imposed by the pace of technological adoption. Its difficulty is political: the beneficiaries of such a fund are precisely the economies competing with Singapore and Malaysia for certain foreign investment. Regional cooperation on this point would require that the best-prepared countries have an interest in their neighbours being prepared too, which is true over the long term, less obvious over the short term.

The institutional window closes faster than employment statistics

The absence of widespread disruption in current employment figures is good news, but misleading good news if read as a guarantee. Employment statistics are lagging indicators: they record effects of decisions made two or three years earlier. Institutions, by contrast, are built over decades.

This lag is the heart of the problem. When net job losses become visible in the data, the moment to act on institutions, negotiation, training, certification will already have passed. Countries that use the period of apparent calm to build these mechanisms will have a lasting advantage. Those that wait for statistical proof of disruption to react will find institutional construction timelines incompatible with the speed of the shock.

The question that remains open, which neither the ILO nor the ADBI resolves, is whether regional governments have the political incentives to invest in institutions whose beneficial effects will be measurable in ten years and whose political costs are immediate. This is probably where, more than in the algorithms, the next decade of work in Southeast Asia will be decided.

Sources

  1. International Labour Organization, Generative AI and labour markets in ASEAN: Significant exposure, limited disruption so far, 8 July 2026
  2. Asian Development Bank Institute (ADBI), Managing the AI Transition, October 2026 (no link: URL not verified)
  3. Technical Education and Skills Development Authority (TESDA), Philippines, digital micro-certifications programme, 2025 (no link: URL not verified)