The AI Farmer’s Friend, How Sarlaben, Vistaar, and Shared Digital Infrastructure Are Transforming Indian Agriculture
Last week, a new voice began speaking to farmers across Gujarat. Her name is Sarlaben, and she is an AI-powered digital assistant created by Amul, India’s iconic dairy cooperative. She answers queries on dairy farming, animal husbandry, and milk procurement in real time. She speaks Gujarati and understands local dialects. She is accessible through the Amul Farmer mobile app, but also via voice calls on basic feature phones—the kind that are ubiquitous in rural India. She will serve over 3.6 million milk producers, most of them women, across more than 18,500 villages.
The launch of Sarlaben is not just a technological achievement; it is a landmark in the diffusion of artificial intelligence in India. As Shankar Murawada’s analysis reveals, the challenges of building such a system are immense. It must work amid intermittent connectivity, understand the nuances of local speech, and provide advice that farmers can stake their livelihoods on. It draws on over 50 years of verified cooperative data, including 2 billion milk procurement transactions annually, veterinary treatment records of 30 million cattle, and farmer-wise cattle census data.
But perhaps the most remarkable aspect of Sarlaben is how quickly it was deployed: just three weeks from concept to launch. This speed was not accidental. It was made possible by a growing ecosystem of shared digital infrastructure, reusable technical architectures, and deployment playbooks that are transforming how AI reaches India’s farmers.
The Challenge: AI at the Real Frontier
Most AI pilots succeed in controlled environments. They have clean data, engaged users, and vendor support on hand. The problems begin when these pilots are scaled. A farming advisory that works perfectly with 500 farmers during a pilot may fail when expanded to 50,000. Farmers find it hard to connect during the rainy season, when they need advice most. The system struggles with unfamiliar dialects. Recommendations conflict with those from local agricultural universities. Slowly, farmers stop using it.
These failures are visible in India precisely because the scale required leaves no room for workarounds. A service that must operate in 22 official languages, account for seasonality, and function with intermittent electricity will reveal every infrastructure gap. A Marathi-speaking farmer using a chatbot on a basic feature phone represents the actual AI frontier—far from the controlled demo environment where everything works perfectly.
The challenge, then, is not just to build AI systems, but to build systems that people can actually adopt in real conditions. This requires a different approach: one that focuses on creating reusable pathways for deployment.
The Pathway Approach: From Nine Months to Three Weeks
Murawada introduces the concept of an “adoption pathway”—a reusable route that combines technical architecture, data and safety governance, evaluation benchmarks, and deployment playbooks. Once a pathway is built, it can be maintained and adapted by others, saving time and reducing risk.
The contrast between three deployments illustrates the power of this approach. Maharashtra’s Vistaar agricultural advisory system took nine months from commitment to deployment in 2025. The state wanted AI-powered advice for farmers in Marathi and local dialects, accessible via basic phones even when connectivity drops during the monsoon. It was the first of its kind, so everything had to be built from scratch: the technical architecture, the governance frameworks, the evaluation protocols.
Ethiopia’s OpenAgriNet took just three months to deploy earlier this year. It addressed the same core challenge—agricultural advice at scale—but it had a pathway already mapped by Maharashtra’s experience. The Ethiopian team could adapt what had already been built, rather than starting from zero.
Amul’s Sarlaben took just three weeks. By the time Amul launched its assistant, the pathway was well-established. The technical architecture, governance frameworks, and deployment playbooks that took nine months to build for Maharashtra could be reused and adapted. The time compression was dramatic.
The Foundation: Shared Digital Infrastructure
This pattern of reuse and adaptation is only possible because of investments in shared digital infrastructure that began years ago. Murawada points to the work of AI4Bharat at IIT Madras, which spent years collecting speech data across 400 districts to build datasets that reflect India’s actual linguistic reality. This foundation enabled the government’s Bhashini language platform, which now serves countless users, and EkStep Foundation’s AXL, which personalizes learning for millions of students in government schools.
These systems have moved beyond pilot projects to become reliable production infrastructure, serving populations larger than many countries. Vistaar provides agricultural advice in Marathi and local dialects because it can draw on Bhashini and AI4Bharat’s multilingual models. Without these shared building blocks, every agricultural system would need to rebuild language capabilities from scratch. With them, deployment becomes repeatable, costs drop, and timelines compress.
This is the opposite of the “not invented here” syndrome that plagues so many technology projects. It is a recognition that some problems—like understanding the diversity of Indian languages—are too large and complex for any single organization to solve alone. By investing in shared infrastructure, India is creating a foundation on which countless applications can be built.
The Trust Factor: Earning Farmers’ Confidence
The technical challenges of building AI for Indian agriculture are immense, but they may not be the hardest part. Murawada suggests that the real friction in adoption may have more to do with risk exposure than technical capability.
Institutions hesitate to adopt AI systems because adoption could fail publicly. It could disrupt existing workflows. It could create compliance burdens. And, most critically, it could create accountability gaps if recommendations go wrong. If an AI assistant advises a farmer to use a certain treatment and the farmer’s cattle die, who is responsible? The farmer? The cooperative? The technology provider?
These are not theoretical concerns. Farmers stake their livelihoods on the advice they receive. An AI system that gives bad advice is not just an inconvenience; it is a potential catastrophe. Building trust requires not just technical accuracy, but clear governance frameworks that establish accountability and provide recourse when things go wrong.
Sarlaben draws on over 50 years of verified cooperative data. This is not just a technical asset; it is a trust asset. Farmers know Amul. They have relationships with the cooperative. The AI assistant is an extension of that trusted institution, not a replacement for it. This institutional grounding is crucial for adoption.
The Women’s Dimension: Serving the Invisible Majority
Murawada notes that most of Amul’s 3.6 million milk producers are women. This is typical of dairy farming in India, which is often women’s work—milking, feeding, caring for the animals. Yet women farmers are often invisible in agricultural extension services, which have traditionally been designed for and targeted at men.
An AI assistant accessible by voice on a basic feature phone has the potential to reach these women in ways that traditional services cannot. It does not require literacy. It does not require travel to a distant office. It is available when and where they need it. If designed with their needs in mind, it could be transformative.
But this requires intentionality. The system must understand the questions women ask, the constraints they face, and the decisions they make. It must be tested with women users and refined based on their feedback. The technology itself is neutral; its impact depends on how it is deployed.
Conclusion: The AI Frontier Is in the Village
Sarlaben, Vistaar, and OpenAgriNet represent a new model for AI diffusion. They are not flashy demos or pilot projects. They are production systems serving millions of users in real conditions. They work in local languages, on basic phones, with intermittent connectivity. They draw on shared infrastructure that took years to build. They are deployed using reusable pathways that compress timelines from years to weeks.
This is the real AI frontier. It is not in Silicon Valley labs or elite research conferences. It is in the villages of Gujarat, the farms of Maharashtra, and the fields of Ethiopia. It is in the voice of Sarlaben, speaking Gujarati to a woman with a basic phone and a buffalo to milk.
The challenges remain immense. Trust must be earned. Governance must be clear. Accountability must be established. But the direction is clear. India is building AI that works for its people, at scale, in the conditions they actually face. That is an achievement worth celebrating.
Q&A: Unpacking AI Diffusion in Indian Agriculture
Q1: What is Sarlaben, and why is its launch significant?
A: Sarlaben is an AI-powered digital assistant launched by Amul for dairy farmers in Gujarat. It answers queries on dairy farming, animal husbandry, and milk procurement in real time, accessible via the Amul Farmer app and voice calls on basic feature phones. It is significant because it demonstrates how AI can be deployed at massive scale (serving 3.6 million farmers across 18,500 villages) in real-world conditions—local languages, intermittent connectivity, and diverse dialects. Its rapid three-week deployment also shows the power of reusable digital infrastructure.
Q2: What are “adoption pathways,” and why do they matter?
A: Adoption pathways are reusable routes that combine technical architecture, data and safety governance, evaluation benchmarks, and deployment playbooks. Once a pathway is built for one project, it can be maintained and adapted by others, saving time and reducing risk. The concept explains why Maharashtra’s Vistaar took nine months to deploy (as the pioneer), Ethiopia’s OpenAgriNet took three months, and Amul’s Sarlaben took just three weeks. Each successive deployment could build on the work of its predecessors, compressing timelines dramatically.
Q3: What role does shared digital infrastructure play in AI diffusion?
A: Shared digital infrastructure—like AI4Bharat’s speech datasets and the government’s Bhashini language platform—provides foundational building blocks that any organization can use. Without these, every AI project would need to rebuild language capabilities from scratch, a prohibitively expensive and time-consuming process. With them, deployment becomes repeatable, costs drop, and timelines compress. This infrastructure reflects a recognition that some problems (like understanding India’s linguistic diversity) are too large for any single organization to solve alone.
Q4: What are the biggest barriers to AI adoption in agriculture, beyond technical challenges?
A: The biggest barriers may be institutional and psychological: risk exposure and trust. Institutions fear that adoption could fail publicly, disrupt workflows, create compliance burdens, or lead to accountability gaps if recommendations go wrong. Farmers, whose livelihoods depend on the advice they receive, must trust the system. Building trust requires not just technical accuracy, but clear governance frameworks, established accountability, and institutional grounding (like Amul’s decades of trusted relationships with farmers).
Q5: Why is the focus on women farmers particularly important?
A: Most of Amul’s 3.6 million milk producers are women, reflecting the reality that dairy farming in India is often women’s work. Yet women farmers are frequently invisible in traditional agricultural extension services, which have been designed for men. An AI assistant accessible by voice on basic phones has the potential to reach women in ways traditional services cannot. However, this requires intentional design—understanding women’s questions, constraints, and decisions, and testing with women users. The technology is neutral; its impact depends on how it is deployed and who it is designed to serve.
