Why one of the year’s biggest AI-for-development stories isn’t actually about AI
What India’s monsoon breakthrough reveals about how technology really works in development
For the first time in the 150-year history of monsoon forecasting in India, millions of farmers received reliable, four-week-ahead predictions of when sustained rains would actually arrive. When the monsoon stalled unexpectedly for nearly three weeks—an event no conventional model caught—the AI forecast did. The Indian government pushed updated guidance to farmers via SMS and voice messages. Farmers who received these forecasts adjusted their planting decisions, and preliminary surveys suggest that around a quarter of the 38 million farmers changed their plans based on the information.
By now, this story has been widely celebrated as a landmark “AI for development” success. And rightly so. In a moment dominated by anxious hand-wringing over whether AI will help or hurt poor countries, this felt like a rare, concrete win.
But as with all good success stories, the most important question isn’t whether it worked. It’s why it worked, and what lessons we take from it.
And here’s the thing: the real story here isn’t really about AI at all.
Don’t get me wrong, the AI models (Google’s NeuralGCM and the European Centre for Medium-range Weather Forecasts’ AIFS) are a remarkable scientific advance. They delivered results that conventional physics-based models simply couldn’t. But if that’s the only lesson we take from this, we’ve learned exactly the wrong thing. We’ll be stuck in the same tired cycle that has plagued development technology for decades: solutions in search of problems, supply-side narratives, and a leapfrogging mentality that treats poor countries as blank canvases for grand experiments.
The India monsoon trial is not a “just add AI” success story. It worked because of a specific constellation of conditions, institutions, and relationships that had been built deliberately over years and even decades, long before AI entered the picture.
What AI did was amplify an opportunity created by a highly specific ecosystem, including:
A concrete, high-stakes decision point (when to plant)
Decades of meteorological data and scientific capacity, including the India Meteorological Department’s century-long rainfall records
Maturing AI weather models, rigorously benchmarked and building on a deep foundation of weather prediction science and practice
A dense institutional network (Indian ministries, global research teams, funders, and practitioners) coordinated around a focused use case
Trusted delivery infrastructure, notably the mKisan platform, launched in 2013 and already reaching tens of millions of farmers in local languages
Iterative, human-centered communication, with messages tested and refined directly with farmers
Strip all these elements away, and what remains is an abstract AI use case that could theoretically help millions of Indian farmers, but unlikely to materialize in practice. And this, in many ways, is precisely where so much of the AI-for development discourse is stuck right now.
The conventional story goes something like this: “AI is transforming everything! Let’s bring it to the poor farmers who need it!” This is supply-side thinking—a hammer looking for a nail. It’s the same logic that has driven two decades of largely failed mobile technology interventions. We keep throwing digital tools at development challenges and wondering why they don’t stick.
The India monsoon trial worked for the opposite reason: it was demand-driven. Indian farmers already relied on and wanted better monsoon forecasts. The government already had infrastructure to deliver agricultural advisories at scale. Global and local researchers already had deep experience with the technically demanding work of monsoon and weather forecasting. And new AI methods happened to offer a technical solution to a well-understood, widely-recognized problem.
The technology was inserted into an existing ecosystem of people, institutions, problems, and opportunities. It amplified what was already there. It didn’t create something from nothing.
There’s a line from The Atlantic’s Charlie Warzel that’s been rattling around in my head lately: “Silicon Valley is not selling useful; it’s selling transformation.” If AI is going to matter for development, we’d be wise to focus more on the useful.
The India AI monsoon trial is a genuine success story, and we should celebrate it. But we should celebrate the right things.
We should celebrate the Indian government’s decade-plus investment in agricultural advisory infrastructure. We should celebrate the meteorological departments that collected and curated a century of rainfall data. We should celebrate the researchers who rigorously evaluated models and corrected for local biases. We should celebrate the message designers who worked with farmers to create understandable, actionable communications. We should celebrate the farmers themselves—who trusted the forecasts, adjusted their decisions, and ultimately made the information matter.
And yes, we should celebrate the AI models that made more accurate forecasts possible.
The question isn’t whether AI can help development—of course it can. The question is whether we’re willing to do the hard, boring, essential work of building the systems that let technology be useful.
The India monsoon trial succeeded because someone did the boring work. That’s the real story. And that’s the lesson worth learning.