Agriculture is perhaps the last valiant sector that refuses to kowtow to the juggernaut of digitization. Medicine got medical records, clinical trials, and PubMed. Law got case archives and statutes. Finance got trading data and filings. Engineering got CAD files, simulations, and patents.
And Agriculture?
Field extension manuals that nobody reads, public data that is brazenly manipulated based on the policy that is being professed, clueless machine-generated data sets that are blithely ignorant about Ashby’s Law of Requisite Variety1and an extension officer chain so thin that one officer covers a thousand farm households in better-served Indian states and far fewer in poorer ones.
The non-generalizable knowledge that actually runs farms is oral, local, deeply contextual and embodied.
A grower who grew up on her land knows where the water clogs after a heavy monsoon, which slopes drain fastest, what the mango orchard needs in its third year versus its seventh. An elephant visiting the farm in the elephant corridor has better prediction engines that tells him when the jackfruit is going to be ripe for illegal consumption. A grower knows that "jilli" in her dialect of Marathi refers to a caterpillar pest at a specific lifecycle stage on a cotton crop, and that the word means something different in a soybean context three districts away.
None of this is on the internet. None of it is in any training corpus. It was passed on, generation to generation, working and traversing the same land together, day in, day out.
This chain is now breaking in large parts of the world, whether in large holding contexts like US or smallholding contexts in India. The next generation does not want to farm. The knowledge is not being transferred.
“Can AI deliver better advice to farmers?” is not the important question. “What is the AI advice drawn from?” is.
When the knowledge that matters most was never digitized, what exactly is the model retrieving when a farmer asks it something?
Last week, I spoke with Two Desais, Sachi and Pratik to delve deep into these questions and more.
Sachi Desai has spent years at the intersection of large-scale precision agriculture and technology at Climate Corporation and Bayer. Pratik Desai is the Founder of KissanAI and Dhenu model. Both offered two complementary answers from their respective contexts that helped us go deeper down the rabbit hole.
Sachi comes from a world where the information gap is less about farmers not knowing things and more about farmers wanting confirmation before taking high-stakes decisions. A soy farmer in Illinois calls her advisor not because she is uninformed, but because farming is capital-intensive and irreversible, and talking through a decision is how she builds confidence to act.
Pratik comes from a world where neither the extension officer, nor the model is present in any meaningful way. When KissanAI trained the first version of Dhenu in 2023, they used approximately 1.5 million farmer conversations as training data because the way a smallholder farmer phrases an agricultural question is almost nothing like the way it appears in any text online. We are dealing with insane amount of context-gleaning skills here.
A generic large language model can only give a satisfactory answer to someone who does not know better. A farmer who knows cotton will immediately identify where the answer falls apart.
The model has not been trained on how farmers speak. It has been trained on how agronomists publish. Can you imagine how divergent the answers could get?
To discover the pathway of convergence, perhaps, its important to peel the business model layer that underpins these systems.
Both Sachi and Pratik operate on a B2B logic: AI advisory tools deployed to agribusinesses, input companies, and retailers who then surface them to farmers. The farmer interacts with an AI that has been configured, constrained, and calibrated by a business whose commercial interest is not identical to the farmer's.
I am turning on the paywall here as we get into the more juicy details.
If you don’t want CC hassles, you are most welcome to use paypal or UPI (venkat.raman.kr@icici) and pay the annual subscription (8500 INR/95 USD) with your email in the comment. I will enable access immediately.
P.S. Supporting this work doesn’t have to come out of your pocket. If you read this as part of your professional development, you can use this email template to request reimbursement for your subscription.
