I spent a morning last August walking a field with a man named Raj. His grandparents planted the first cotton on this land in 1952. His daughter, a software engineer in Bengaluru, had recently convinced him to try a soil-moisture sensor network and a little AI assistant that lives in an app on his phone.
Raj is in his sixties. He was, as he put it, "skeptical but not a fool."
What I saw that morning taught me more about introducing AI into long-standing work than a dozen boardroom conversations ever have.
The thing that worked
The system Raj uses doesn't try to be clever. It takes readings from a scattering of moisture and temperature sensors, cross-references them with weather forecasts and his planting calendar, and sends him one short message a day:
"Water field 3 this evening. Hold field 7 — rain likely Thursday."
That's it. No dashboard. No "insights." Just a sentence.
He reads it with his morning tea. Sometimes he agrees, sometimes he doesn't — his own instincts, fifty years deep, are still the final word. But the sensors have paid for themselves twice over in the first season, mostly by stopping him watering fields that didn't need it.
The thing that almost didn't
The first version of the app, his daughter told me, was much fancier. Charts. Predictions. A "crop health score." Raj tolerated it for a week and then quietly stopped opening it.
"Too many numbers," he said. "I know my fields."
The second version — the one that works — is a dramatic downgrade in technical ambition and a dramatic upgrade in usefulness. The cleverness is in what gets left out.
A few things we've learned
Working with agricultural clients, a pattern keeps showing up:
- Decision support, not decision replacement. The farmer is the expert. The AI is the junior colleague with better eyesight and a longer memory.
- Small sensor networks beat ambitious satellite pipelines. For most farms, a dozen well-placed sensors and a simple model outperform a fancy computer-vision system analysing satellite imagery — and they're easier to fix when something breaks.
- Offline-first is non-negotiable. Rural connectivity is patchy. An AI tool that stops working when the 4G drops is a tool that gets uninstalled.
- One useful alert a day builds trust. Seventeen builds resentment.
Where it's genuinely helping
- A dairy co-op in the Netherlands using a vision model to spot the early signs of lameness in cows — days before a human can see it — so the animal gets treated before it suffers.
- Smallholder coffee farmers in Colombia using a phone app to identify leaf rust from a photograph, then getting an agronomist's WhatsApp number if they want a second opinion.
- A wine estate in France using AI-assisted forecasts to time harvests, which, as the winemaker told me, is "like having a very clever apprentice who doesn't drink."
None of these are futuristic. None of them replace the human. All of them pay their keep.
What I'd say to anyone building in this space
Go and spend a day doing the job. Not watching — doing. Walk the fields. Help with a harvest. Stand in the milking parlour at 5am. You'll learn more in one morning than a year of user interviews, and whatever you build afterwards will be twice as useful and half as arrogant.
Raj's text message came in while I was writing this. "Field 9 needs water. Also, thank you for the visit — my wife says come back and eat with us." That's the version of AI-in-farming I want to live in.
If you're working on something similar, drop us a line. We're always up for a farm-gate conversation.
