In Bangladesh, the manufacturing sector employs approximately 10% of the active population. When Taiwan and South Korea were going through their industrial takeoff, at a comparable income level, industry’s share of total employment was already heading toward 25 to 30%. The staircase still exists. It has simply become too narrow to accommodate tens of millions of people.

This is the finding that Dani Rodrik, Harvard economist and one of the most listened-to voices on development, published in May 2026 in Project Syndicate. What makes this text unusual: Rodrik is not commenting on others’ mistakes. He is publicly documenting why his own central thesis, defended for a decade, no longer holds up against the data.

The Essential

  • Bangladesh, Vietnam, and Cambodia structurally absorb less labor in industry than Taiwan, Korea, or Japan did at an equivalent development stage — a gap that Rodrik documents in his thesis revision (Project Syndicate, May 2026).
  • Automation is shrinking the absorption capacity of manufacturing before poor countries have even had time to fully exploit it.
  • Rodrik now proposes betting on certain services — logistics, healthcare, tourism, digital — as a possible substitute, but the data on this bet remains preliminary.
  • If the transition to services fails to create mobility at large scale, low-income countries in sub-Saharan Africa and South Asia still betting on industrialization could miss their window, with consequences for the trajectory of several billion people by 2040.

What Rodrik Defended, and Why It Worked

Rodrik’s manufacturing thesis rested on a solid observation: the industrial sector is one of the few to offer rapid productivity gains in poor countries without requiring high human capital at entry. A textile worker in Bangladesh can multiply her productivity by five or ten by moving from subsistence agriculture to the factory. These gains are transmitted to wages, local services, and domestic demand. South Korea, Taiwan, Japan, and China followed this path, and the data is unambiguous: hundreds of millions of people escaped poverty through this door.

What the theory captured well was the mechanism. Manufacturing shares a characteristic that few other sectors possess: unconditional convergence of productivity. Manufacturing firms in poor countries tend to catch up to the global technological frontier, regardless of institutions, education levels, or governance quality. In other words, the poorest countries could still climb the staircase, provided they had access to export markets and global value chains. This is what Rodrik called “industrial globalization” as an engine of catch-up.

The problem is that this convergence property applied to a manufacturing model that absorbed labor. When automation enters the equation, productivity continues to converge, but employment no longer follows.

The Staircase Narrows Under Automation

China is the most striking illustration of this mechanism. Between 1995 and 2015, Chinese industry absorbed hundreds of millions of rural workers. Then, around 2012, Chinese manufacturing employment began to decline in absolute terms, even as industrial production continued to grow. Robots and automation decoupled production from labor. What was happening in China was not an anomaly: it was a sign that the model was changing in nature.

For countries arriving later on this staircase, the figures Rodrik documents are revealing. Vietnam, which experienced rapid industrialization in the 2010s thanks to relocations fleeing rising Chinese costs, today employs approximately 16% of its active population in manufacturing. This is significant, but well below the 30 to 35% that Taiwan or Korea reached at a comparable income level in the 1970s and 1980s. Cambodia and Bangladesh present similar or even lower ratios. The factories are there. The jobs per unit of production are much fewer.

This phenomenon has a name in economic literature: “premature deindustrialization.” Rodrik himself had theorized it in earlier work, but he saw it then as a partial phenomenon, potentially compensable through appropriate policies. What he acknowledges in 2026 is that the phenomenon is deeper and more structural than he anticipated. Automation does not only reduce the cheap labor advantage of poor countries: it reduces the volume of employment that industry can physically create, even in the least-skilled sectors.

The connection to institutional dynamics is direct. As we documented regarding Mexican nearshoring, comparative advantage is no longer enough: countries that attracted factories through low wages see this advantage erode as technology reduces the weight of wage costs in the production structure. When wages represent only 8% of the total cost of an assembled product, competition over labor changes in nature.

Services: A Reasonable Bet, Thin Data Still

Rodrik is not arguing for abandoning industrial policies. He contends that the services sector can fulfill, under certain conditions, a function similar to what manufacturing played: creating economic mobility at large scale for low-skilled workers.

The distinction he introduces is important. Not all services are equal. Low-end personal services (cleaning, childcare, small food service) do not offer the productivity gains that fuel catch-up. In contrast, certain segments present interesting properties: tourism, logistics, low-tech healthcare, and a fraction of digital services accessible to semi-skilled workers. These are the segments Rodrik identifies as candidates for a new engine of mobility.

The argument holds, in part. Costa Rica built significant economic upgrading through export services, notably medical devices and shared service centers. Mauritius followed a comparable path. Several sub-Saharan African countries see their informal services sector becoming a net absorber of urban employment. These examples exist. What they do not yet prove is that the model is scalable to the same extent as manufacturing.

The major difference, and this is the weak point of the argument, is that services do not spontaneously possess that property of unconditional convergence that Rodrik attributed to industry. Productivity gains in services are more dependent on institutions, governance, education levels, and connectivity. A poor country with deficient institutions can still set up a competitive textile factory; building an ecosystem of competitive exportable services without a solid institutional foundation is substantially more difficult. Rodrik acknowledges this limitation, but he does not yet have a satisfactory answer to the question: how do countries with weak institutional capacities access this new model?

The Liberal Tension: When Meritocracy Becomes Structurally Difficult

This is where Rodrik’s reading enters into tension with another school of thought. Nicolas Bouzou, a liberal economist and careful observer of growth dynamics, has consistently argued that pessimism about development often reflects a cultural pathology rather than economic diagnosis: institutions, economic freedom, commercial openness remain the most reliable engines of catch-up, provided countries believe in them and build them. Optimism is a productive resource.

This reading has real force. Economies that have managed to escape poverty over the past fifty years did not do so despite openness, but through it. And it is true that catastrophism about premature deindustrialization risks leading to counterproductive protectionist policies or demobilization of local actors.

But the tension Rodrik surfaces is more precise than that. It is not about the value of openness or institutions. It is about absorption capacity. If manufacturing can accommodate 10% of the active population where it once accommodated 30% a generation earlier, countries with strong demographic growth, particularly in sub-Saharan Africa, face an arithmetic problem that neither solid institutions nor liberal optimism alone resolves. They would need to create tens of millions of productive jobs in sectors where data on absorption capacity remain preliminary.

Lucid optimism consists of holding both readings together: yes, services can become a new engine, and certain countries prove it. And, simultaneously, the model is not yet documented at the scale that African or South Asian demographic transition demands. The question is not resolved. Intellectualizing it is not either.

The Long Arc: A Closing Window

Demography makes this issue urgent. Sub-Saharan Africa is projected to add approximately 800 million people to its population by 2050, according to UN projections. Most will enter the labor market in economies that have not yet completed their industrialization and that face increasing automation in low-value-added manufacturing sectors. The same pressure, at a lesser scale, affects Bangladesh, Pakistan, and several South Asian countries.

What was a problem of economic policy becomes a question of historical window. Countries that industrialized between 1960 and 1990 benefited from a specific context: limited automation, low communication costs, open export markets, and manufacturing technology that required labor. This context will not repeat for countries arriving now. If services do not fulfill the role Rodrik assigns them, large-scale social mobility for this generation will be slower, more unequal, and probably insufficient to absorb demographic pressure.

There is something analogous in this finding to the question posed by data on AI and the transformation of cognitive work in advanced economies: automation does not necessarily destroy jobs on net, but it profoundly modifies the structure of who accesses what, and how quickly. The difference is that rich countries have safety nets, redistributive institutions, and human capital to manage this transition. Most low-income countries do not have these buffers.

The partial irreversibility of this phenomenon is what makes it serious. Global value chains will likely reconfigure toward more automation and nearshoring to high-cost countries as robots become cheaper. This trend is documented by the OECD and various sectoral studies: automation costs in textiles and electronics assembly fell approximately 40 to 50% over the 2012-2022 decade. Each additional point of decline reduces the comparative advantage of cheap labor in emerging countries.

What Countries Are Already Doing, and What It Says About Current Bets

Several countries have begun building concrete responses. Rwanda has bet on positioning itself for high-end services: conservation tourism, international conference hub, regional logistics. Results are visible in terms of GDP per capita and formal employment, but the economy remains small and the model difficult to replicate at the scale of a country with 50 or 100 million inhabitants. Kenya is developing a digital ecosystem around Nairobi that employs hundreds of thousands in outsourced services and fintech. India, at another scale, has built an export services industry that now represents more than 8% of its GDP and employs several million qualified workers, before diffusing toward less-skilled segments.

These examples share a characteristic: they all benefited from significant public investment in digital infrastructure, active training policy, and openness to foreign investment in services. In other words, the mechanisms that allow services to fulfill industry’s role resemble active industrial policy more than spontaneous market growth. Rodrik, consistent with his earlier work on the developmental state, draws this conclusion explicitly.

This is practical information for governments and international institutions. The World Bank and development agencies continue to direct a significant share of their financing toward manufacturing infrastructure and special economic zones. If Rodrik’s diagnosis is correct, part of these resources should migrate toward training in services, digital infrastructure, and regulatory frameworks that allow small service providers to participate in global value chains.

AI could play an ambiguous role here. On one hand, low-cost accessible digital tools lower the entry barrier into certain exportable services: an accountant in Ghana can now work for a European SME with tools her predecessor could not have afforded. On the other hand, the same tools automate the service segments most accessible to low-skilled workers. The question of which of these two effects will dominate remains open, and available data at this stage do not allow for resolution.

An Economist Updating His Map

What gives value to Rodrik’s text, beyond its content, is the method it illustrates. Changing one’s mind based on new data is rarer than one might think in economic debate. Most economists defend their earlier theses by adding nuance, not by acknowledging that the reference model was correct at one moment and ceased to be. Rodrik does the opposite. He says: the world has changed, my analytical framework must change too.

This posture has a practical consequence for institutions that finance development and for governments that depend on them. Development strategies from the 2000s and 2010s, built on the model of special economic zones and integration into global assembly chains, did not fail: they worked in the context that was theirs. That context has changed fast enough that the same recipes, applied today in Sudan, Ethiopia, or Myanmar, would produce structurally weaker results.

The task that opens up is to build a theory of development through services as precise and as operational as the manufacturing theory was. Rodrik acknowledges that this work is not finished. The conditions under which services can fulfill this function, the policies that make them accessible to low-skilled workers, the institutional models that enable this upgrading: these are the questions that will need to be resolved empirically, country by country, sector by sector.

The staircase has not disappeared. It has changed location. And no one quite has the map of the new building yet.

Sources

  1. Dani Rodrik, “The New Path Out of Poverty”, Project Syndicate, May 2026 — https://www.project-syndicate.org/commentary/services-not-manufacturing-best-hope-for-developing-countries-by-dani-rodrik-2026-05
  2. Dani Rodrik, “Premature Deindustrialization”, Journal of Economic Growth, 2016 (earlier work on premature deindustrialization, no guaranteed URL)
  3. United Nations, World Population Prospects 2024, demographic projections for sub-Saharan Africa
  4. OECD, sectoral studies on automation costs in textiles and electronics, 2022-2024
  5. Journal d’un Progressiste, “Le nearshoring mexicain recule, l’avantage compétitif ne vaut rien sans les institutions” — https://journaldunprogressiste.fr/le-nearshoring-mexicain-recule-faute-dinstitutions-lavantage-competitif-ne-vaut-rien-sans-les-institutions/
  6. Journal d’un Progressiste, “Krugman and Aghion: How economists can read the same data and conclude the opposite?” — https://journaldunprogressiste.fr/krugman-et-aghion-lisent-les-memes-donnees-et-concluent-linverse/