AI / Feb 2026
AI in extension services: hype, reality and a path forward
Where ML genuinely helps smallholder farmers — and where simpler tools still win.
By Ivan Rono
Every agritech pitch deck now mentions AI. Some of it is real. Most of it is a wrapper around a chatbot that doesn't know your soil. Here's an honest look at where machine learning actually moves the needle for smallholder farmers.
Where AI genuinely helps
Possible applications include image-assisted crop diagnostics, forecasting and voice interfaces. Evaluate each against local data, realistic devices and a clear baseline. Keep uncertainty visible and involve agricultural experts in validation.
Where it doesn't (yet)
Generic LLM chatbots giving agronomy advice are dangerous. They hallucinate spray schedules, mix up local cultivars, and miss agro-ecological context. A decision tree reviewed by an agronomist can provide a more verifiable starting point.
A practical path forward
Start with a narrow, verifiable model — one crop, one region, one decision. Pair it with a human extension officer who can override. Measure adoption and yield, not engagement. The boring playbook beats the demo every time.