In a recent commentary, Dani Rodrik, one of the world’s leading experts in trade and development economics, explains why he became a “manufacturing skeptic” after decades of viewing industrial production as the key to unlocking economic growth. Since traditional manufacturing cannot absorb the 1.5 billion workers “in occupations that do not require a college education or are exposed to the international economy through trade or offshoring,” service-driven growth models must take center stage. If opportunities in retail, hospitality and food services expand, middle-class consumption will drive productivity gains.
But largely missing from this analysis is the potential for human productivity gains augmented by artificial intelligence. You only have to look around the world to find that AI is already quietly increasing productivity in labor-intensive services in ways that do not require or presuppose a college education. This trend is especially evident in India, which offers a way to reconcile “manufacturing skepticism” with the persistent desire for industrialization of policymakers in Africa and South and Southeast Asia.
India’s services-led growth reflects more than an expansion of its retail and hospitality sectors. Algorithmic tools have fundamentally transformed operations through inventory optimization, dynamic pricing, demand forecasting, and supply chain coordination. As a result, a formal retail operation in Bengaluru or Mumbai today is fundamentally different from a store in the 1990s, not because the employees are more educated, but because algorithms have made them more productive.
The same dynamic can be found even among street vendors and other micro-enterprises. With platforms like Flipkart B2B and JioMart offering informal traders AI-powered tools for demand forecasting and purchasing optimization, a small vegetable seller with just a mobile device can anticipate what customers will want, source accordingly, and make margins that were not available just five years ago.
Consider a parallel to another moment in industrial history. In the 1960s, a textile worker in South Korea probably produced 50 times more cloth than his Indian counterpart, not because he was more capable, but because the loom was better. Now, AI does the same for street vendors, retail workers, and small farmers. The big difference is that these tools can be made much more accessible. They are much easier to implement than factories and work with existing informal structures, rather than requiring a complete industrial reorganization.
One implication of this difference is that it may not matter that 1.5 billion workers remain in low-skilled, non-tradable occupations. Even if general occupational patterns persist, productivity can grow. We already know that a smallholder farmer equipped with a digital agronomic advisor in East Africa or South Asia can achieve average productivity gains of 30% without additional land or formal education. A vegetable seller who uses simple AI tools to optimize inventory and prices can double her income in 18 months.
The main limitation to productivity growth is not technological, but institutional. Putting new tools in the hands of workers requires localization in native languages, accessibility without the need for advanced digital literacy, sustainable business models with micro margins, and integration with informal financial systems.
Crucially, policy makers must also ensure that these systems are built on an inclusive open access architecture. If dominant e-commerce platforms use predatory fees to appropriate the excess value generated by inventory optimization, the peddler’s hard-earned margins will simply flow onto corporate balance sheets. Meeting all of these requirements may not be easy, but it is possible.
Similar findings apply to manufacturing. While competing for a place in modern global manufacturing value chains requires sophisticated skills, AI is fundamentally altering the equation. A garment factory no longer needs 500 multi-skilled workers; It needs 50 workers complemented by quality control systems using computer vision, demand-driven production planning and logistics optimization. Operations of this type can already be found in Bangladesh and Vietnam.
Of course, skeptics will point to the immediate displacement of the remaining 450 workers. But the broader economic transformation that will be underway should not be overlooked. As hyperproductive factories scale up production and reduce unit costs, they catalyze a vast ecosystem of jobs outside the factory, from upstream supply chain coordination to downstream distribution, which can reabsorb labor at scale.
The productivity paradox that Rodrik identifies—the fact that the expansion of manufacturing employment no longer goes hand in hand with productivity growth—is real, but not insoluble. When small informal manufacturers gain access to AI-enabled quality checks, supply chain coordination, and just-in-time planning, they can achieve levels of productivity previously only available to large integrated factories.
Mexico’s weak economic performance under NAFTA supports this argument. The problem is not that Mexican manufacturers could not compete, but that technological diffusion failed to reach small and informal operations. Factories expanded, but without the right tools and platform technologies, workers remained stuck in low-margin assembly tasks.
The same technologies can reduce human capital requirements in many other sectors. While it is true that India owes much of its success in service-led growth to its large English-speaking population and early investments in information technology infrastructure, similar models can be replicated elsewhere. Thanks to advances in AI translation, English-speaking engineers are not needed to deploy digital agriculture tools in places where people speak Amharic, Swahili or Telugu.
Therefore, whether service-led growth is more feasible than industrialization in poor countries is a secondary question. Viability depends solely on the availability of mechanisms that improve productivity. In the 20th century, that meant access to industrial machinery. But today it means access to AI tools that operate in local languages, work offline, and cost just pennies per transaction.
Of course, widespread use of these tools will not happen automatically. Most investment in AI today is concentrated in high-skill, high-wage economies. But this reflects political decisions, not some economic law. The question for policymakers is whether they will treat localized, affordable AI for informal workers as infrastructure — like rural electrification or highway systems — or as a luxury good for the rich.
If developing countries invest in making AI and related digital tools as widely available as possible, manufacturing-led growth and services-led growth can coexist. A farmer who uses AI to improve his yields can achieve middle-class income without leaving farming, and a street vendor can achieve the same without joining a large organized company.
Manufacturing does not need to be the engine of growth; it simply needs to coexist with formal and informal services powered by AI. This vision may be less romantic than the East Asian industrialization narrative, but it has the advantage of being achievable right now. The main challenge is to implement a deliberate and inclusive infrastructure investment strategy that facilitates the adoption of AI. The best way to elevate those 1.5 billion people is to meet them where they are.
Ravi Venkatesan, president of the Global Energy Alliance for People and Planet, sits on the boards of Hitachi and ServiceNow.
