A small exporter in a developing economy can now use artificial intelligence to translate a product description, compare foreign regulations and draft a message to a potential buyer. The cost of getting started has fallen. The cost of a wrong answer has not.
That distinction should guide the next stage of international economic policy. Governments are understandably eager to build data centers, secure access to advanced models and attract AI investment. But the most valuable question for many economies is whether local firms and workers can use the technology to enter new markets, negotiate on better terms and retain the expertise that lets them compete after the software changes.
The World Trade Organization’s World Trade Report 2026, published in September, describes how digitalization, artificial intelligence and global value chains are testing a trading system built for an earlier era. It also observes that low- and middle-income economies increased their share of world trade from 23 percent in 1995 to 45 percent in 2024. Those countries have a much larger stake in the rules that shape the next technological transition.
The World Bank’s World Development Report 2026, published in August, argues that developing economies can benefit by adopting available AI, adapting it to local conditions and gradually advancing their own capabilities. That sequence matters because building the largest general-purpose models is financially unrealistic for most countries. A government that treats frontier-model ownership as the only serious AI policy may spend heavily while neglecting affordable improvements available to ordinary businesses.
Yet adoption alone also has limits. Imagine a small supplier using AI to generate bids for international contracts. The system may shorten a day’s paperwork to an hour. If the supplier cannot verify a tariff classification, interpret a foreign quality standard or detect a fraudulent payment instruction, the gain in speed becomes a new vulnerability. The firm needs people who can check claims, document decisions and escalate uncertainty.
For trade and development policymakers, I would propose a skills dividend attached to major AI investments. When a government subsidizes digital infrastructure or negotiates a technology partnership, it should define what local workers and enterprises will learn to do independently. The deliverable would be measurable competence in commercially important tasks, not a tally of new accounts or training certificates.
The first priority should be export-facing small and medium-sized enterprises. Public trade agencies, chambers of commerce and development banks could offer short practical clinics built around real documents: a customer inquiry, a customs requirement, a purchase order and a dispute about delivery terms. Participants would use an approved AI system to draft a response, then demonstrate which requirements they verified from original sources.
Such clinics would reveal where AI adds value and where it creates rework. An exporter might produce quotations faster while repeatedly misunderstanding a documentary requirement. A training provider should measure corrected submissions, successful handoffs to human experts and the time until a buyer receives reliable information. These results provide a better basis for investment than the number of prompts used.
A second priority is local adaptation. The World Bank emphasizes access to tools that work with local languages and the constraints of basic phones, intermittent connectivity and uneven technical capacity. Policymakers should test products with the communities expected to use them. An agricultural cooperative, a small manufacturing supplier and a rural medical clinic have different informational needs and different consequences when an AI answer is wrong.
Trade agreements and investment memoranda could support interoperability, the ability to switch providers without losing essential records, and fair access for smaller domestic firms. Those provisions would help preserve competition and local bargaining power. The objective need not be a single international rulebook for every AI use. It can begin with clear expectations about the portability of work products, the ownership of decision records and the right to question automated recommendations.
There is also a workforce dimension that trade ministries cannot hand entirely to technology vendors. The International Monetary Fund’s September 2026 annual report discussion of AI highlights both the investment surge and the exposure of middle-skilled workers to automation. A country can attract AI-enabled service exports yet discover that employers have removed the junior assignments through which employees once learned the work.
Governments should therefore help firms protect a ladder of responsibility. Entry-level staff can begin by checking AI-assisted outputs against source documents, then progress to explaining exceptions, supervising customer communication and eventually designing better procedures. Managers should record what new workers can do after three months that they could not do at the beginning. Otherwise, a firm may become fast at producing routine text while gradually losing the people who know when that text is misleading.
This is especially important in outsourced business services, an industry in which cross-border competition often rewards speed and price. A buyer in a wealthier country may see cheaper customer support or faster document processing. A supplier in a developing country should be able to show something more durable: fewer corrections, dependable escalation, stronger worker skills and better service to the buyer’s own customers.
Countries should pursue different combinations of policies according to their infrastructure and institutions. A well-connected service-export economy may focus on worker progression and advanced quality assurance. An economy with unreliable electricity may obtain greater immediate benefit from simple offline-capable tools and basic connectivity. The common standard should be demonstrable local value, measured in the environment where the technology will operate.
International partners can play a constructive role here. Development lenders could finance training and evaluation alongside equipment. Trade capacity-building programs could publish reusable assessment exercises. Importing firms could reward suppliers that document how they verify AI-assisted work, rather than squeezing all gains into lower contract prices. None of those steps requires a new treaty before a pilot can begin.
The case for AI-enabled trade remains strong. The WTO sees considerable potential for digital technologies to reduce barriers and expand services trade, while the World Bank identifies practical opportunities to amplify the capabilities of workers in developing economies. The policy failure would come from confusing access to technology with the ability to use it well.
The next round of AI diplomacy should ask three concrete questions. Which local firms will be able to compete for new business? Which workers will gain skills that remain useful when a tool or vendor changes? What evidence will show that the gains survived the first inaccurate answer, disputed invoice or interrupted connection? A credible AI partnership should give those questions answers before the celebratory announcement. Infrastructure attracts headlines. A local workforce that can make reliable decisions will determine who keeps the economic gains.
.svg.avif){kind=small}
