Iese

Will AI upend the offshoring of global knowledge work?

The globalization trend of outsourcing routine service jobs to emerging economies may get reversed, depending on which way GenAI develops and grows.

May 26, 2026

During the globalization and ICT revolution of the ’80s and ’90s, the world of knowledge work diverged into two blocs: Advanced economies specialized in complex problem-solving services, while emerging economies focused on routine knowledge tasks. The classic example is of multinationals offshoring their call centers to countries with English-speaking, low-cost workforces, such as the Philippines and India, while R&D labs and higher-end sales-and-services divisions stayed in the United States.

The spread of generative AI (GenAI) has the potential to reverse these decades-long divisions of labor, shifting the patterns of imports and exports of services, an increasingly large part of global trade. But will it? Or will past trade patterns be reinforced in the AI era? And what will determine which way things go?

Advanced problem solvers vs. routine knowledge workers

AI’s impact on international trade depends on the level of the technology’s problem-solving capabilities, driven by computational power. Crucially, AI infrastructure, such as data centers, is concentrated in advanced economies, especially the United States, giving them unique control over computational power.

If AI agents possess basic knowledge, equivalent to standard pre-AI workers, emerging economies can incorporate AI agents as additional workers, and the most knowledgeable and competent human employees advance into higher-skilled, problem-solving roles. The emerging economy becomes less specialized in routine knowledge work, and, if the shift is large enough, could even begin to export problem-solving services. In this scenario, pre-AI trade patterns would potentially be reversed.

If, on the other hand, AI agents possess the sophisticated knowledge of higher-level pre-AI employees, the advanced economy’s specialization in complex problem solving deepens. The supply of problem solvers (both human and AI) would grow in advanced economies, with the computational infrastructure to support them. The growing numbers of advanced problem solvers would increase demand for routine knowledge work done in emerging economies. In this scenario, existing trade patterns would be reinforced.

Trade dynamic mirrors within-firm trends

These potential transnational trends build on findings of AI’s impact on junior- and senior-level employees within firms. Early research had suggested that GenAI would help junior employees the most, because it serves as a useful assistant that narrowed the knowledge gap between entry-level workers and their managers. In this line of thinking, AI would augment the work of novice employees.

But as AI becomes more autonomous, the opposite may turn out to be true: The least knowledgeable workers might be substituted by AI, while the most knowledgeable workers may always benefit, since they can pair their expertise with ever-more sophisticated AI.

The determining factor is autonomy — the possibility that AI agents can be given tasks that take hours, days or weeks to complete, and then they can go off and do those tasks autonomously, in the way a smart employee would.

With autonomous AI, the most knowledgeable workers move into specialized problem solving and use AI as an autonomous agent. But the least knowledgeable workers may be substituted or, if not substituted, they will be left with the least successful managers because the best ones will be using AI agents.

MORE INFO:

“The impact of AI on global knowledge work” by Enrique Ide and Eduard Talamas is published in the Journal of Monetary Economics (2026).

“Artificial intelligence in the knowledge economy” by Enrique Ide and Eduard Talamas is published in the Journal of Political Economy (2025). This paper received a 2025 Research Excellence Award, granted by the IESE Alumni Association.

This article is included in the annual publication, Insight for Global Leaders No. 2 (2026).