The Middle-Income Trap Has Two Growth Engines, and the Handoff Between Them Isn't Automatic

A country that spends three decades getting cheap at making things is not, by that fact alone, equipped to get clever at making things. Economists studying the middle-income trap increasingly frame it less as a single barrier and more as a handoff between two different growth engines, one built on labor costs and one built on ideas, that does not happen on its own. Research on deindustrialization, technology absorption, and automation points to where that handoff typically breaks, and only a handful of middle-income economies complete that climb.

Manufacturing's On-Ramp Is Closing Earlier Than It Used To

Cheap labor works as a growth strategy because manufacturing can absorb large numbers of workers directly out of subsistence agriculture and pay them more for it, while the sector is unusually good at raising their productivity through learning by doing and exposure to competitive export markets. Economist Dani Rodrik has documented that the relationship between a country's income and the share of its workforce in manufacturing follows a hump shape: employment in the sector rises through early industrialization, peaks, then declines as an economy matures. What has changed, per Rodrik's analysis, is where that peak now falls. In recent decades, manufacturing's employment and output shares have started declining at income levels and industrial shares well below what earlier industrializers reached before their own peaks; developing economies are running out of industrialization opportunities sooner, and at lower income levels, than the historical pattern would predict.

That timing matters because manufacturing is not one sector among many in this framework. It is the sector most directly tied to the productivity gains a knowledge economy needs to inherit. When its employment share peaks early, workers who would once have kept moving into higher-paying industrial jobs are pushed instead into low-productivity informal services, and the country loses years it could have used to build the human capital, institutions, and firm capabilities the next stage requires.

The World Bank's own account of the transition, developed for its 2024 World Development Report, maps this same handoff onto three sequential strategies and shows roughly where the deindustrialization risk above sits inside that sequence.

Imported Technology Only Works If Someone Can Absorb It

Per the World Bank's World Development Report 2024, the strategy that works for the poorest economies, investment in basic infrastructure and factories, has to give way to a different strategy once wages start climbing: infusion, meaning the deliberate import and domestic diffusion of technology, management practices, and business models developed elsewhere. The report describes this as powered mainly by flows of physical and financial capital, most visibly foreign direct investment. A separate strand of the growth literature, built around the concept of absorptive capacity in technology transfer, complicates that picture: importing a technology and actually raising domestic productivity with it are different events. Absorptive capacity, the ability to acquire, adapt, and apply an imported technology rather than simply operate it, depends heavily on a country's existing stock of human capital and firm-level capability. A factory can install imported machinery without any of that capacity spreading into the rest of the economy.

That is why the World Bank's own prescriptions for the infusion stage lean so heavily on the workforce side of the ledger: broadening foundational skills, improving learning outcomes, and building links between local and globally leading universities sit alongside the trade and investment measures. Infusion is a capital-flow strategy that pays off only if a human-capital strategy runs alongside it.

Set side by side, the two growth engines this trap sits between rely on close to opposite inputs, which is part of why the transition between them is not a smooth continuation of the same policies that got a country to middle income in the first place.

Two World Bank Reports Don't Fully Agree on Automation's Timeline

A 2020 paper by economists Linda Glawe and Helmut Wagner, available as an open-access working-paper version ahead of its appearance in Asian Economic Papers, argues that automation and artificial intelligence sharpen the absorptive-capacity bottleneck rather than ease it. Their concern is that the frontier technology associated with the current wave of automation is more skill-biased than earlier generations of imported machinery, so the workforce entering the infusion stage today needs a higher floor of technical skill than the workforce that carried earlier economies through the same stage.

World Bank researchers who published the 2026 World Development Report on artificial intelligence reach a more measured conclusion than that mismatch argument. Jobs in high-income countries are more than three times as likely to face automation risk from generative AI as jobs in low- and middle-income countries, largely because more work in those economies is manual rather than cognitive. On the same data, the report estimates AI could lift productivity in more than 16 percent of existing jobs in developing economies, against more than 18 percent in advanced ones, a gap narrow enough that near-term disruption looks smaller in developing economies than near-term opportunity does.

That finding sits in tension with the mismatch argument above, not in agreement with it. Glawe and Wagner treat automation as raising the skill floor a workforce needs before it can even enter the infusion stage. The World Bank's 2026 report treats the same technology as something a workforce can adopt with comparatively little near-term displacement, provided the surrounding conditions (power, connectivity, skills, institutional quality) are in place. Both claims can hold together only if the two reports are describing different time horizons: WDR2026 measures which jobs are exposed to automation today, while Glawe and Wagner describe what happens once automation-driven capital deepens further and the frontier technology available for infusion keeps climbing in skill-intensity. The World Bank's own report leans toward that later-stage concern in its own framing: its recommended path is to adopt, then adapt, then advance, and advancing is explicitly conditioned on the same skills and institutional-quality gaps Glawe and Wagner flag, just measured against a later point in AI adoption than today's.

The mechanism echoes at the individual level too, in ways reporting on Stanford's labor-market data has already documented: entry-level and routine roles are the ones most exposed to automation over time, and they are also the roles a labor-intensive economy uses to absorb workers moving out of agriculture in the first place. If that first rung thins as adoption deepens, the labor-intensive growth engine has less runway before it needs the knowledge-economy engine to take over, even as that second engine's own entry threshold keeps rising.

What the 3i Model Doesn't Settle: Sequencing Under a Shrinking Middle Stage

The World Bank's 2024 framework treats investment, infusion, and innovation as a sequence, one strategy giving way cleanly to the next as a country crosses each income threshold. The deindustrialization and automation findings above sit uneasily with that assumption, because both suggest the middle stage is being compressed from two directions at once: manufacturing's capacity to absorb labor now peaks earlier, and the technology available to infuse now demands more skill than an economy has typically had time to build. Neither the deindustrialization literature nor the World Bank's own two most recent flagship reports offer a settled answer to what happens when a sequence's stages arrive faster than the human capital and institutions underneath them can be built. The 2024 report describes stages separated by income thresholds; the 2026 report suggests automation pressure on those stages arrives on a timeline of its own, not tied neatly to GNI per capita bands at all. That is not a gap better country-level statistics would close on their own. It is a genuinely open question in the growth literature about whether a strategy designed around sequential stages still describes the transition when the middle stage keeps getting shorter.