By Kalpana Kochhar
Many countries are trying to do more for their citizens, albeit often with outdated systems. That means accessing government services can feel like a second job. A mother walks half a day to a clinic, only to be turned away because she needs a form from an office in another town.
A farmer who qualifies for financial support gives up after being passed from queue to queue. AI, if used correctly, could help close these gaps, expanding a capable state’s reach. For low- and middle-income countries, the stakes could not be higher: AI could be a great equalizer, or it could deepen inequality.
While AI’s likely contributions to agriculture, health care, and education hold great promise, the World Bank’s World Development Report 2026 estimates that AI’s productivity benefi ts could be more than twice as large in advanced economies if current trends continue. The technology’s future trajectory depends on the decisions that developers, investors, and governments make today about how it is built, funded, and deployed.
As Kristalina Georgieva, managing director of the International Monetary Fund, explained at the IMF-World Bank Annual Meetings in October, AI adoption depends on electricity access, digital connectivity, and workers equipped with the right skills. But after decades spent in development finance, I am convinced that access, while a legitimate concern, is not the biggest barrier.
Above all, AI adoption requires building the state capacity to put these tools to use. In advanced economies, the AI debate has largely focused on safety, security, and the threat of job displacement. But in developing economies, the more immediate question is how to ensure that AI tools actually serve people.
They must work in local languages, train on local data, and be built for local realities. And they must be coupled with investment in the institutions and people closest to the problem, who understand where AI tools can deliver real benefits and can ensure that they do. Much of this essential work falls to the companies building these tools and the funders backing them.
But even the best-built tools will struggle to deliver results if governments lack the capacity to implement them. Creating that capacity requires integrating AI systems into existing programmes. Consider MahaVISTAAR, a free AI advisory service that the Indian state of Maharashtra built for farmers to ask about crops, weather, and pests in their own language through an app, voice call, or chat.
So far, more than 740,000 farmers have joined, with 70,000 new users signing up each month, at a cost to the government of less than 18 cents per farmer—a fraction of what it would cost to reach them in person. The tool worked because the government viewed it as a new layer of public infrastructure built on its network for advising farmers, rather than a one-off app.
On a recent visit to Paki stan, I witnessed a provincial government working to build this kind of capacity, which it views as a public good. Punjab has established a government office dedicated to AI and is collaborating with local researchers to develop public-health applications, including local-language tools that help frontline workers reach mothers with the right guidance at the right time. AI tools intended to serve citizens must work within government, not merely be adapted to it.
They must be built to run on public systems and data, adhere to the data-sharing and security protocols that policymakers have selected for their countries, and be affordable enough for governments to procure and sustain them. In my experience, three elements are essential to building AI capacity.
First, governments need leaders who can determine the best use cases for AI, alongside institutional ownership and cross-ministry coordination. An AI strategy needs a clear budget and detailed timelines, and should be backed by empowered ministries and a mechanism that makes decisions binding across agencies.
This could reduce the likelihood of parallel projects, which may result in duplication and suboptimal outcomes, eroding the limited capability that already exists. Second, countries need civil servants with enough AI expertise to commission, acquire, integrate, evaluate, and regulate these systems. That means developing talent pipelines that channel people with the necessary technical skills—distinct from general digital literacy—toward government roles. While not even wealthy countries have figured out how to meet this demand, developing countries can hardly afford to neglect it.
Third, governments must build integrated digital public infrastructure, including e-government service delivery, digital ID, payments, and a secure data exchange layer. Weak digitisation and fragmented vertical systems are a direct constraint on AI performance and safety. Once that infrastructure is in place, AI can in turn help make government systems more secure—detecting fraud, cyberattacks, and anomalous patterns.
Building the tools may be the more straightforward part. Whether AI reaches that mother or that farmer will be determined less by the pace of the technology than by the choices governments make now. If policymakers get it right, AI can help a country fulfil its most basic, and most important, duty: serving its citizens effectively.
The writer is Interim President of Global Policy & Advocacy at the Gates Foundation.
