When Automation Creates Jobs—and Still Displaces Workers
Industrial robot adoption in Viet Nam is the subject of this series. The central question is not whether robots are, in the abstract, good or bad for employment; it is how the same wave of automation can expand jobs and wages in some parts of a local economy while removing formal work from people whose tasks are easiest to automate. Understanding that split gives readers a more useful way to judge claims about automation: ask what happened to tasks, output, workers, and places—not only whether the total number of jobs went up.
The evidence comes from a July 2026 IZA Discussion Paper, Industrial Automation and Jobs in an Emerging Economy: Evidence from Viet Nam. The paper studies differences across Vietnamese districts, firms, and worker groups during a period when industrial robot adoption accelerated. It offers a sharply defined case, not a universal forecast. Viet Nam's experience is valuable precisely because it puts technological change inside an emerging economy's development model, where export manufacturing, foreign investment, labor mobility, and large differences in skills operate at the same time.
Viet Nam's automation episode is a development story, not just a technology story
For decades, emerging economies have often used abundant labor and labor-intensive manufacturing as a route into industrialization. Viet Nam followed that path while deepening its integration into world trade. Reforms, industrial parks, foreign investment, and access to global value chains helped the country become an important export platform. As the composition of exports shifted toward electronics and machinery, the production system became a setting in which robots could do more than replace a worker at one workstation: they could also help firms meet demanding quality, cost, and volume requirements in international markets.
The speed of the change matters. The paper reports that Viet Nam's industrial robot density—the stock of robots per thousand manufacturing workers—grew more than thirtyfold between 2010 and 2022. By 2022, the country was near Malaysia and Thailand at roughly eight robots per thousand manufacturing workers. Adoption accelerated especially after 2015 and was concentrated in electrical equipment, computers and electronics, automotive production, and rubber and plastics. These were not random pockets of activity. They were industries connected to export markets and to the intermediate inputs needed by more technologically advanced production.
The paper's four leading robot-adopting industries accounted for 32.5% of total Vietnamese exports in 2015 and 41.5% in 2022. That figure does not mean that all exports were automated, nor that every export-sector job depended on a robot. It identifies the economic environment in which the measured automation occurred. If a firm introduces a robot while serving a fixed local market, the immediate labor-saving effect may dominate. If the same firm becomes more productive, lowers costs, expands its foreign sales, and increases production, the market around the robot can grow. The labor-market result then depends on the balance between those forces.
The unit of observation is also important. The study does not ask whether Viet Nam's entire economy had more jobs after automation than it would have had in some perfectly isolated alternative history. It compares districts with different initial industrial structures and therefore different exposure to the industries that later adopted robots more rapidly. A district with a large pre-existing electronics or electrical-equipment workforce was positioned differently from one whose employment was concentrated elsewhere. That local comparison allows the paper to examine how automation was absorbed by particular labor markets, while leaving open questions about national welfare and long-run transformation.
The practical stakes follow from this setting. A district can become more productive and attract new investment while workers within it experience very different futures. A skilled technician may find a better-paid formal job. A worker whose routine manual task has been automated may move into informal work, leave the district, or compete for a narrower set of jobs. New demand for food, transport, construction, and services may create opportunities that do not resemble the factory jobs that disappeared. Calling the result simply “job creation” would miss those transitions; calling it simply “job destruction” would miss the expansion that makes some of them possible.
The same robot can cut a task and expand a market
The useful starting point is the task, not the occupation. A robot may perform material handling, welding, assembly, cutting, or another repeatable production activity without replacing every responsibility associated with an occupation. The direct substitution effect is nevertheless real: when a machine performs a task previously carried out by workers, firms need fewer hours of that kind for each unit of output. The workers most exposed are likely to be those whose formal jobs are built around routine manual activities.
That is only one side of the mechanism. Automation can also reduce unit costs, improve consistency, and make it profitable to produce more. Lower costs may allow a Vietnamese exporter to compete for additional orders abroad. Higher output can increase demand for workers who install, supervise, maintain, coordinate, inspect, or manage the new production system. It can also generate work in activities that are not themselves automated. Suppliers need logistics and business services; workers with higher incomes buy local goods; expanding industrial areas need construction and other non-tradable services.
These channels do not automatically cancel each other. Their relative force depends on the demand for the product, the firm's ability to expand, the skills available to complement the technology, and the extent to which inputs are imported rather than purchased locally. A productivity gain that mainly replaces labor in a stagnant market has a different employment consequence from a gain that helps an export-oriented firm capture a rapidly growing market. Viet Nam's trade integration therefore belongs inside the explanation, not in a separate background section.
The same logic explains why an average district result cannot be read as a statement about every worker. Scale expansion may increase the number of jobs while changing their skill requirements. A firm may employ more people overall and still need fewer workers for a particular routine task. The additional jobs may be located in foreign-owned enterprises, in complementary occupations, or in services around the industrial hub. They may be available to new migrants more readily than to older residents whose experience is tied to the displaced work.
This distinction also changes how to interpret wages. When productivity rises, firms may pay more for workers whose skills complement the new capital. But workers in tasks that have become less valuable may face weaker bargaining power or a move into informal work, where the paper finds no comparable wage gain. A district-wide average can rise even while the bottom of the distribution is under pressure. Aggregate improvement and unequal adjustment are not competing descriptions if they refer to different workers and different channels.
What the evidence must keep separate
The paper's headline results make the puzzle concrete. In its preferred local estimates, an additional robot per thousand workers is associated with a roughly 6.5–9.7% increase in total employment and about a 3–5% increase in average wages for formal workers. Yet the employment rate does not significantly rise. The number of people and the number of jobs grow together, a pattern consistent with the robot-exposed districts receiving more workers. Formal and informal salaried employment effects are generally positive but not statistically significant, while non-salaried employment increases in specifications that exclude the most highly exposed districts.
Those averages sit beside a more uneven pattern. High-skilled employment rises, including in formal work, while formal wage employment among low-skilled workers falls. The paper locates the losses especially among routine manual occupations and finds higher informality among low-skilled workers performing those tasks. It also finds that employment shifts toward foreign-owned enterprises and away from domestic private firms. The initial lesson is therefore conditional: the exposed districts appear to gain economic activity, but the gains are not distributed in proportion to the losses created by task substitution.
The study addresses the risk that more successful districts simply adopted more robots for reasons unrelated to the technology. It constructs district exposure from each district's 2011 industry employment shares and later industry-level robot adoption. It then uses robotization trends in the same industries across 56 other countries as an instrument for Vietnamese exposure. District fixed effects, regional year effects, baseline characteristics, and Chinese-trade controls are used to address other differences. Pre-trend checks are broadly reassuring after controls, although high-exposure districts already had stronger economic conditions and some differential pre-trends that require care.
This design supports a particular kind of conclusion. It estimates relative local effects associated with differential robot exposure during the main 2014–2020 estimation period. It does not count every robot physically installed in every district, and it does not directly model all migration, commuting, trade, or wage equalization across the country. The labor-force survey includes informal and self-employed workers but did not collect self-employed earnings during the study period. The enterprise survey covers registered firms and excludes informal firms and self-employment. The two sources therefore illuminate different parts of the adjustment rather than providing interchangeable measurements.
The timing also sets a boundary. The study excludes 2021 and 2022 from its core labor-market estimation because the pandemic disrupted economic activity and data collection. Its illustrative calculations of approximately 255,000 formal jobs for workers with tertiary education and 66,800 displaced formal jobs among low-skilled workers extend estimated effects through 2022 under stated assumptions. They are local-equilibrium calculations relative to average exposure, with wide confidence intervals, not a precise national ledger of jobs caused by robots.
The most important conclusion at this stage is methodological as well as economic. To understand what automation did in Viet Nam, the series must keep four questions separate: which tasks were substituted; whether firms used productivity gains to expand output; which workers could access the complementary jobs; and where displaced or newly recruited workers went. The next step is to examine those forces in sequence, beginning with the task-level mechanism that makes employment creation and displacement possible at the same time.