FarmOS • Mobile App • Concept
FarmOS
FarmOS is a concept product for pineapple farmers working plots around Vazhakulam, Kerala. Instead of another farm management dashboard, it puts an AI agent at the center: one that opens the morning, closes out the evening, and coordinates with workers and vendors on the farmer's behalf, with every outgoing call or message confirmed before it happens.
- Engagement
- Self-directed concept project
- Duration
- 1 Week
- Scope
- Product concept · mobile UX · agentic conversation design
- Year
- 2025
Most farm management software assumes the farmer wants a system to learn. FarmOS tests what happens if they don't have to.
Fields, plots, crop cycles, task lists and vendor contacts, each filed into its own screen, waiting for someone to go find them. A pineapple farmer working plots around Vazhakulam, Kerala doesn't think in database tables. They think in what needs doing today, and who they need to call.
FarmOS is a concept product built around that distinction: an AI agent that understands the farm, not an app the farmer has to understand.
This is self-directed work, not a client engagement. We built it to test an approach, and to have a real answer ready the next time a client asks whether AI belongs at the center of a product instead of bolted onto the side.
Edwin is a native of Vazhakulam and spoke with three to four local pineapple farmers directly. These are informal conversations, not a formal study, but they're where the decisions below actually came from.
Informal conversations, not a formal research study. Small sample, one region.
Coordination, not record-keeping
Every farmer described the same pattern: plots, workers, vendors and weather tracked in their head and in phone calls, not in any app. Nobody asked for a place to log data. They wanted fewer calls to make.
Hands are busy
Typing detailed updates mid-task came up as a real friction point, not a hypothetical one. Farmers described wanting to talk to something, not type into it, while out in the field.
Trust is earned action by action
When asked about an AI making calls or sending messages on their behalf, the hesitation wasn't about the AI being wrong. It was about not knowing what it had done. That shaped the confirmation model more than anything else in these conversations.
The direction flipped once, and the flip mattered more than any single screen.
The earliest version looked like most farm software: a dashboard home screen with farm stats up top and a chat icon tucked into a corner. That fell apart on its own logic. If the agent was supposed to be the product, burying it behind a stats screen contradicted the premise.
The second shift was separating "the agent talks" from "the agent acts." Early versions had the agent narrate what it was doing in the same breath as doing it. Splitting the confirmation moment, Connect versus I'll Handle it Myself, from the action itself made it clear when FarmOS was about to leave the app and contact a real person on the farmer's behalf. That's the one place a silent agent would have broken trust fastest, and it's what the Vazhakulam conversations pointed at directly.
This is concept work, not a deployed product, so the honest next step is testing it against more farmers before any of this gets treated as settled: whether a slash-command brief is the right entry point or should just run on its own, how much a farmer actually trusts an agent to make a call on their behalf, whether voice changes the task-creation flow enough to redesign around it, and real vendor and worker data beyond the handful of conversations this was built on.
Broken doesn't mean beaten. Untested doesn't mean wrong.