If artificial intelligence is said to affect millions of Thai workers, it is easy to picture software replacing people on a large scale. The latest evidence, however, points to a more complicated transition.
Thailand's National Economic and Social Development Council (NESDC) applied a generative AI exposure framework to labour-force survey data from the fourth quarter of 2025. It estimated that about 8.7 million workers, or 21.8% of Thailand's labour force, were in occupations potentially affected by generative AI.
The same estimate contains an important distinction. About 2.2 million workers, or 5.4% of the labour force, were in a task-replacement group. A much larger 6.5 million, or 16.3%, were in a task-augmentation group where AI is more likely to support productivity while core duties still depend on human skills.
That makes the more useful question not simply "How many jobs will AI take?" but which parts of work will AI change, and where will human value move after those parts become cheaper or faster to perform?
AI exposure does not automatically mean job loss
A job is usually a bundle of activities. Some can be automated, some can be accelerated, and others still depend heavily on communication, negotiation, judgment or responsibility.
NESDC separates these effects into two broad groups. Task Replacement covers work where AI can substitute for part or all of structured, repeatable tasks. The group is estimated at about 2.2 million workers and includes parts of clerical, accounting and programming work.
The larger Task Augmentation group, estimated at 6.5 million workers, includes occupations where AI can handle some steps or process information more efficiently while important duties still depend on people.
The regional evidence points in a similar direction. The International Labour Organization estimated that 22.9% of ASEAN employment in 2025 was in occupations with more than minimal exposure to generative AI, while only 3.3% was in the highest-exposure category. As of July 2026, the ILO reported little evidence of widespread labour-market disruption from generative AI.
That does not make the risk irrelevant. It means "work is changing" and "jobs are disappearing" should not be treated as the same statement.
AI often reaches the task before it reaches the whole job
Consider a finance employee who gathers data, builds tables, checks anomalies, explains findings to management and stands behind a recommendation. Or a marketer who researches customers, drafts material, evaluates campaigns, speaks with clients and decides where to allocate a budget.
AI may become much faster at some of those activities without eliminating the value of every other part of the role.
If AI can produce a first draft in minutes, the value of producing that first draft may fall. But the value of knowing which data to trust, what looks wrong, whether a conclusion is too risky, and what decision should follow may become more important.
This is why occupational exposure alone can be misleading. A highly exposed occupation can contain both tasks that are vulnerable to automation and tasks where AI raises productivity, meaning the amount of useful output produced from a given amount of resources.
When a task becomes cheaper, firms have more than one option
Suppose software turns an hour of work into fifteen minutes. The cost of that task has fallen, but the firm's next move is not predetermined.
It could reduce headcount. It could also keep the same team and produce more, serve more customers, test products faster, or redirect staff toward activities that require more judgment. If output per worker rises and the business can sell the additional output, AI can support expansion rather than immediate job contraction.
The opposite is also possible. If demand does not grow and little valuable work remains after automation, firms have a stronger incentive to reduce positions. Employment outcomes therefore depend on technology, business models, market demand and workers' ability to move toward tasks where AI remains weaker.
There is no simple equation in which a 20% productivity improvement must produce a 20% reduction in employment.
The work that becomes more valuable may be the work that requires judgment
As tools become better at structured production, the scarce capability may shift away from producing text or calculations quickly and toward understanding context, asking the right question and taking responsibility for the outcome.
The OECD reports that AI adoption is raising the skills bar in many workplaces. More than half of surveyed employers in manufacturing and finance reported a greater need for highly educated workers after adopting AI. At the same time, the OECD estimates that fewer than 1% of workers will need advanced AI-specific skills such as model development or AI programming.
For most workers, the relevant skill set is broader: using, analysing and interpreting data, together with problem-solving, creativity, management and other human skills.
If AI makes the "first answer" cheaper, workers may create more value by being able to frame the problem, verify the answer, connect it to business context and make decisions when the information is incomplete.
Those skills are not guaranteed to remain protected from automation. But during a period of organisational transition, they are often what turns AI from an experiment into an operational result.
The more immediate risk may be the first rung of the career ladder
Labour-market adjustment does not have to arrive as a wave of layoffs. It can also appear as fewer openings for new workers.
Entry-level jobs have traditionally included tasks such as gathering information, drafting documents, checking files, conducting basic analysis and solving structured problems. Those activities also provide the practice through which junior workers accumulate experience before moving into roles that require more judgment.
If generative AI reduces the need for junior staff to perform those tasks, the first rung of some career ladders could narrow even if experienced workers remain employed.
US data from the Stanford Digital Economy Lab is useful as a warning signal, not as evidence that the same pattern already exists in Thailand. Payroll data through June 2026 showed that employment among 22 – 25 year-olds in highly AI-exposed occupations stood about 19% below where it would have been if it had kept pace with employment among similarly aged workers in less-exposed occupations. The adjustment appeared primarily through weaker hiring rather than higher separations.
Stanford explicitly describes these as descriptive patterns, not causal estimates of AI's effect. Other labour-market forces may also matter. Even so, the finding raises an important question for other economies: if AI performs more junior tasks, how will young workers obtain the experience needed for senior ones?
Not everyone needs to become an AI engineer
The response to AI should not begin with the assumption that every worker needs to learn machine learning or model development.
For many occupations, the advantage may come from understanding the existing profession more deeply, using AI to reduce time spent on repetitive work, and improving the parts that depend on domain knowledge, verification and judgment. An accountant may use AI to organise information but still needs to validate the result and interpret what it means. A marketer may use AI to generate options but still has to understand the customer, the budget, the brand and the business objective.
For employers, buying an AI tool does not automatically create a productivity gain. Weak data, poorly designed processes and inadequate verification can make work faster while also making errors easier to scale.
The economic opportunity therefore lies in redesigning work so that people and AI each perform the parts where they have an advantage, and in turning productivity gains into more valuable products or services rather than simply doing the same work in less time.
What to watch next
The first signal is entry-level hiring in Thailand, especially in office support, finance, accounting, marketing, technology and basic analytical roles. If junior vacancies weaken faster than overall employment, companies may be redesigning the career ladder.
The second is the skills employers request in job postings. The important shift is not simply whether the word AI appears, but whether demand moves toward analysis, verification, communication, project management and industry-specific knowledge.
The third is wages and employment in AI-complemented work compared with work whose tasks are easier to automate. A widening gap would reveal the labour-market effects of AI more directly than the number of tools companies buy.
The final signal is employer investment in upskilling and reskilling, including whether small and medium-sized businesses can adopt AI effectively. If productivity gains are concentrated among large firms, differences in efficiency and wages across businesses could widen.
AI can reduce the value of some tasks. That does not mean it reduces the value of human work at the same rate. The increasingly important question is what happens to the part of the job that remains after AI takes over one layer of it, and whether workers and firms can move toward that value quickly enough.
