Agriculture is usually described through its constraints: falling groundwater, erratic monsoons, small and fragmented holdings and volatile markets. Less often discussed is a structural problem underneath them: the distance between where agricultural intelligence is generated and where decisions are made.
Over the past decade, sensors, satellite imagery, agronomic research, and predictive models have greatly expanded what we can know about crops, soil, and climate. Yet much of that knowledge stays far from the farmer's day. The data exists. It does not arrive at the moment of action, in a form that supports a decision.
A companion piece covers why agricultural AI needs an integrated layer across crops, livestock and markets. This one covers what decides whether any of it reaches the farmer: how the advice is communicated, and how it learns.
Intelligence per acre, not more inputs
The next gains in farm productivity are unlikely to come from more fertiliser or more water. They will come from better-timed, better-judged decisions: when to irrigate, how much to fertilise, whether to harvest now or wait, and where to sell.
Those decisions are local. They depend on soil, crop stage, labour, market access and the household's tolerance for risk. A static advisory or a generic dashboard cannot hold that variety. What is needed is advice that adapts to context: that can work with partial information, ask a clarifying question and turn complex signals into a clear next step.
Extension has always been about conversation
Agricultural extension, the system of advisers who carry research to farms, has always run on conversation. An extension officer walks the field, asks what the farmer has seen, and gives advice in the local language. Its limits are equally old. There are too few advisers for too many farmers, and advice often arrives ’s needed.
Earlier digital efforts showed what helps. India's Kisan Call Centres, running since 2004, give farmers a free number to call experts in their language. Digital Green's research in India found that short videos of local farmers demonstrating practices, discussed in village groups, were far more cost-effective than conventional extension visits. Both point to the same lesson: farmers trust advice delivered in their language, by people like them, when they need it.
Communication is the design problem
Much of AI's success or failure in agriculture depends on a neglected layer: communication. Farmers do not work through spreadsheets or forms. They work through observation, conversation and experience. Systems that demand rigid data entry or technical fluency put the burden on the people least able to carry it.
Conversational interfaces change that. Voice and text, in the farmer's own language, at any hour, let a farmer describe what they see in their own words, share a photograph of a leaf or a pest, and get a clear answer. The system does the work of structuring and interpreting. India's national language programme, Bhashini, and a growing set of Indian-language speech and translation models now make this practical across many of the country's languages.
Good agricultural conversation has particular features:
It asks before it advises: crop stage, recent weather, what the farmer has already tried.
It accepts what farmers send: a voice note, a blurry photograph, a description in local terms.
It is honest about uncertainty: when a photograph is unclear, it says so and asks for another, or refers the farmer to a person.
It respects the season: it raises sowing, irrigation and harvest questions when they matter, not all at once.
It speaks plainly: in the farmer's language and register, without agronomic jargon.
From delivering advice to learning from the field
Traditional advisory treats knowledge as centrally produced and sent outward. Conversational AI enables a different architecture. When farmers describe conditions, answer follow-up questions and report what happened, they generate high-quality signals from the ground.
Over time, those signals add up to a shared understanding of how crops behave in particular places, seasons and practices. The system learns not only from sensors and satellites, but from lived experience. Advice becomes more local and more credible. Learning flows both ways: from research to field, and from field back to research.
Orchestration, not one more platform
India's agricultural ecosystem already contains much of what is needed: soil health records, weather data, remote sensing, mandi prices, logistics and credit. The Digital Agriculture Mission's AgriStack adds farmer registries and crop records. The challenge is not availability but coordination.
Rather than collecting everything into one platform, the more resilient approach is orchestration: connecting existing systems at the point of decision. A farmer asking whether to sell this week needs today's mandi prices, their own stock, the weather outlook and their loan schedule, combined into one answer. No single system holds all four. A conversation can bring them together.
Trust is earned in the field
Farmers have long experience of advice that did not fit their land. A system earns trust the way a good adviser does: by getting the small things right, admitting what it doesn’t know, and bringing in a person when the stakes are high. It should start with advice the farmer can ignore, and take on more only as it proves reliable in that context.
What to measure
How many farmers come back after their first question.
How often advice was followed, and what happened when it was.
Which questions recur by region and season.
How quickly unclear cases reach a human expert.
Whether farmers' own reports improve the advice over time.
How Mitochondria's agentic AI systems support agricultural advisory
The approach. Mitochondria's research with agricultural organisations starts by mapping how farmers ask for and receive help: channels, languages, seasons and the questions that recur. Systems work by voice and WhatsApp in Indian languages, configured to the organisation's agronomy, products and farmer data.
What it enables. Advice at the moment of decision, in the farmer's language, connected to what the organisation already knows about each farm. Every conversation adds to structured knowledge the organisation can use, and field experts join when judgement is needed.
The impact. Farmers get usable advice when it matters, and organisations sharpen their guidance with every season.
Read more about Mitochondria's approach.
Sources
Gandhi, R., Veeraraghavan, R., Toyama, K. and Ramprasad, V. (2009). Digital Green: participatory video and mediated instruction for agricultural extension. Information Technologies and International Development, 5(1).
Government of India, Kisan Call Centre (since 2004).
Government of India, Bhashini (National Language Translation Mission).
Government of India (2024). Digital Agriculture Mission (AgriStack).
—
Mitochondria builds agentic AI for evidence-led core operations. Research-led and model-agnostic. ISO/IEC 27001:2022 certified, with controls aligned to the GDPR, UK GDPR, India's DPDP Act and Australia's Privacy Act. Amsterdam and Pune. Request a working session →