Studies of the impact of robots on labour markets often conceptualise jobs as bundles of distinct tasks. In such frameworks, technologies substitute or complement workers. But this view may overestimate the potential adverse effects of automation, particularly in developing countries. This column distinguishes between technical feasibility and economic viability of adopting industrial robots and artificial intelligence to examine their labour market impact. In several middle- and low-income countries, the labour-saving effects of robots have been offset by export-driven increases in the scale of production. While AI threatens fewer jobs in developing countries, these countries are also less equipped to take advantage of the benefits of AI.
A growing body of literature offers both theoretical and empirical assessments of the impact of automation technologies on employment, typically through the lens of technical feasibility and typically on industrial countries. For instance, studies examining how robots affect labour markets often adopt a task-based framework, which conceptualises jobs as bundles of distinct tasks. According to this perspective, new technologies may substitute for workers in performing routine tasks while complementing them in others (Autor et al. 2003, Acemoglu and Autor 2011, Acemoglu and Restrepo 2019).
However, a limited number of studies argue that an exclusive focus on technical feasibility may overestimate the potential adverse effects of automation – particularly in developing countries (UNCTAD 2017, Artuç et al. 2023). These studies emphasise that what is technically feasible is not always economically profitable.
Our recent work (Arias et al. 2025) contributes to this discourse by introducing an integrated framework to understand how robots and AI affect labour markets. We distinguish between technical feasibility – whether the tasks associated with a job can be performed using new technology – and economic viability – whether it is profitable for firms to adopt the technology, given its costs and benefits. By linking these two dimensions – what can be automated and what makes economic sense to automate – the framework helps explain the heterogeneity in automation across sectors and countries.
Furthermore, to assess the labour market impacts of adopting new technologies, we need to consider two effects. The first is the labour displacement effect of technology arising from the automation of tasks. The second is the productivity effect: automation can lower production costs and prices, leading to higher product demand, greater output, and therefore more jobs.
Whether technology leads to net job losses or gains ultimately depends on the balance between these two effects. If displacement outpaces productivity-driven job creation, overall employment declines. Conversely, if productivity gains and new job creation exceed displacement, employment can rise.
Automation is technically feasible for a growing number of tasks
Figure 1 maps representative occupations in terms of their relative task content to assess the technical feasibility of technology affecting jobs. The analysis builds on the work of Acemoglu and Autor (2011) and Lee (2018). Each occupation is a bundle of tasks, some of which may be performed by machines while others remain with humans.
Figure 1 Task and jobs framework
Notes: Adapted from original frameworks by Acemoglu and Autor (2011) and Lee (2018). AI = artificial intelligence. IoT = internet of things. LLM = large language model.
New technologies impact jobs depending on whether they involve (a) cognitive or manual tasks, and (b) routine or non-routine tasks.
- Earlier digital technologies – including robots in industrial sectors and computer and digital software – automated many routine physical and cognitive tasks, from welding parts on an assembly line to filing paperwork. Thus, they impacted the jobs of assembly-line workers, clerks, and tax preparers (Figure 1, bottom and top-left quadrants).
- The internet and, recently, AI have allowed machines to take on more complex non-routine cognitive tasks, impacting the jobs of financial analysts, risk assessors, and translators (top-right quadrant).
- AI also has the potential to augment humans in jobs involving strategy, creative and social tasks, in service jobs like those of teachers, radiologists, and managers.
- AI-powered devices or smart robots are increasingly able to perform a wider range of non-routine manual tasks such as auto-driving and service tasks, thus potentially affecting taxi and truck drivers as well as restaurant attendants (bottom-right quadrant).
Technology also creates entirely new tasks and roles. For instance, digital technologies have given rise to the jobs of IT support specialists, data analysts, and, more recently, drone operators and AI prompt engineers.
Figure 2 (panels A and B) applies this stylised framework to measures of the task content and the actual occupational structure of emerging and advanced economies. In high-income economies, the largest share of employment is in non-routine cognitive roles (reflecting a services-led economy). Jobs that are both routine and manual – like machine operators or clerical staff – have already experienced significant automation. Low-income countries still have a high share of workers in agriculture and routine manual jobs. Middle-income countries lie in between – outside of agriculture, they tend to have a higher share of routine manual jobs and routine cognitive jobs than either rich or poor countries. These are the jobs most vulnerable to automation.
Figure 2 Task intensity and occupational structures of emerging and advanced economies
A. Emerging markets and developing economies
B. Advanced economies
Notes: Data from Autor and Dorn (2013); ILOSTAT, International Labour Organization; O*NET Online, Occupational Information Network, Employment and Training Administration, US Department of Labour. Routine task intensity (RTI) calculated following Acemoglu and Autor (2011); the reserve RTI value used for horizontal axis. Bubble size represents employment share in each industry. Advanced economies = 36 countries. Emerging markets and developing economies = 76 countries.
Automation is also increasingly economically viable
While many routine tasks are ripe for automation from the technical-feasibility standpoint, the economic incentive to automate depends on robot prices and labour costs. If robots are too expensive or labour is very cheap, firms will not adopt robots.
The cost and economic viability of robots vary across industries and countries. New technologies are most economically viable in high-wage sectors. Focusing on evidence from major East Asian countries, our study finds that robots used in the automotive sector are about 10 times more expensive than those used in lower-wage sectors like rubber and plastics. Thus, in Indonesia and the Philippines, robots are largely utilised in rubber and plastics, while in China, Malaysia, and Thailand, robot adoption extends to electronics and automotive sectors (Figure 3).
Figure 3 Wages and robot adoption across countries/sectors
Notes: Data from OECD, IFR, ILOSTAT. Horizontal axis shows the annual average wage of a worker in $1,000. Vertical axis shows log of change in robot stock per worker (2014–19). In order to include 0 changes, 0.001 is added to the change in robot stock.
Industrial robots have a positive but uneven employment impact in developing countries
Studies on how robotisation affects labour markets have primarily focused on developed countries, largely because robots were predominantly adopted in these economies. Existing empirical evidence indicates that the adoption of industrial robots has hurt aggregate manufacturing employment and wages, especially of low-skilled workers engaged in routine manual tasks (see, for example, Graetz and Michaels 2018, Aghion et al. 2020, Acemoglu and Restrepo 2020). These effects have often compounded the impact of import competition from developing countries.
In contrast, automation has had a positive impact on employment in manufacturing in several developing countries, as demonstrated by our empirical analysis of the major developing East Asian countries. The reason may be the comparative advantage of Southeast Asian countries in manufacturing, because of which firms face a high price elasticity of demand in global markets. This high elasticity means the productivity-driven reductions in costs and prices translate into big export-driven increases in the scale of production, which offset the labour-saving effects of robots. China and Viet Nam, which have seen the most rapid growth in robot penetration, have also seen the faster rise in the share of exports and industrial employment. In Viet Nam, for instance, an additional robot adopted per 1,000 workers raises the exposed district’s total employment and average labour wages by approximately 6%–9% and 2%–4%, respectively.
However, the benefits have not been evenly shared. Between 2018 and 2022, industrial robot adoption created around 2 million jobs for skilled formal workers, but it also displaced 1.4 million low-skilled formal workers, in routine and manual jobs, across the five ASEAN countries studied. The chief beneficiaries have been younger workers, like engineers, equipped with skills that enable them to work with robots. Many of the displaced are older assembly-line workers, who have found refuge in the informal sector. Our causal estimates, presented in Figure 4, suggest that ASEAN-5 locations experiencing greater robot adoption have seen increases in employment and earnings of the higher-educated, but lower average wages of the least-educated workers.
Figure 4 Robot adoption and labour market impacts in ASEAN-5 by skill level
Notes: 2SLS estimates of the effects of exposure to robots on local labour market outcomes in ASEAN-5 countries. Exposure to robots is measured as the interaction between the baseline employment composition by industry in each administrative division and robot adoption by industry-year in each country and is instrumented with ‘global exposure to robots’ that uses the average robot adoption by industry-year across 54 countries (following Acemoglu and Restrepo 2020). Based on data availability, an administrative division is defined at the district level for Indonesia (pre-2015) and Viet Nam, and at the province level for Malaysia, Philippines, and Thailand. Low-skilled: primary education (or below); Middle-skilled: secondary education or high school; High-skilled: vocational, college, or higher education. All regressions weighted by the baseline population in each administrative division and controls for location fixed effects and country × year fixed effects, baseline demographic characteristics in each division (log population; share of urban population; share of migrants; shares of population with primary, secondary, and tertiary education; shares of population under ages 21–55 and older than 56; and share of females), the divisions’ baseline industry shares (employment in primary, manufacturing, services, and the female share of manufacturing employment), and baseline economic characteristics (employment rate, unemployment rate, labour informality rate, share of salaried employment, share of self-employment, female employment rate, exposure to job routinisation, log average hourly wage, log average labour income, and log total labour income).
AI: A broader, but slower-moving wave
It is too early to assess the economic viability and actual rates of AI adoption. Therefore, our analysis is limited to the technical feasibility of AI to displace or augment workers, particularly in jobs in services. Our analysis suggests that the still nascent AI technology has not yet produced noticeable labour market impacts in developing countries.
AI can augment humans in jobs in education and healthcare, and in jobs involving strategy and decision-making. However, only about 10%–12% of jobs in East Asia and the Pacific and other emerging economies involve such tasks complemented by AI, compared to the 30% share in advanced economies.
Moreover, AI exposure is not uniform across workers. In East Asia and the Pacific, women are more exposed than men to the AI-displacement effect. Workers with tertiary education are more exposed to AI displacement than workers with secondary education or less.
Looking ahead, however, AI is likely to transform production processes while digitalisation is enhancing the tradability of services. As occupational structures shift toward cognitive-intensive jobs in services and wages rise, developing economies will not be immune. Indeed, many of today’s ‘safe’ jobs could become tomorrow’s prime targets for AI displacement. Just as export-oriented manufacturing helped shift employment from agriculture to industry in developing East Asian and Pacific countries, these and other emerging economies need to harness the potential of AI to ensure that a dynamic services sector is a source of better job opportunities in the future.
Policies to turn technological advances into a blessing rather than a curse
The evidence from East Asia, a paragon of inclusive labour-intensive development, suggests that emerging economies may be standing at the cusp of a new technology-driven wave of labour market disruption. As new technologies become cheaper and labour costs rise, their adoption is likely to spread and transform labour markets. Understanding how technical feasibility and economic viability interact is critical for crafting policies that maximise the benefits and mitigate the risks of automation and AI.
Policymakers must act proactively to prepare for the transformative impact of automation and digitalisation. Key reforms include investing in skills development to equip workers with advanced technical, digital, and socioemotional competencies – such as critical thinking, creativity, and teamwork – that complement emerging technologies. Countries like Japan, Korea, and Singapore are already advancing these efforts through revamped curricula and digital-learning platforms.
Labour and capital mobility should also be facilitated by removing barriers that hinder worker relocation or sectoral shifts, and by liberalising services sectors to promote trade and competition (de Nicola et al. 2025, Barattieri and Mattoo 2024). Additionally, governments must address factor-price distortions such as capital tax exemptions or labour taxes that led to excessive automation and stunted job growth in advanced economies. Equally important is the expansion of social insurance schemes to support workers in the growing digital informal economy. This expansion can be achieved through better information dissemination, financial incentives, and behavioural nudges, with both public- and private-sector involvement.
Source : VOXeu
