← All food signals
Emerging

The Vernacular Agronomist

A farm-advice chatbot built on farmer questions and expert answers replies in 16 local languages for small farms. Two registered randomised trials are testing what chatbots like it change.

What is it?

AI advice for small farms is being built from farmers’ own questions and delivered in their own languages. Digital Green’s FarmerChat replies in 16 local languages, and two registered trials are testing what such chatbots change. Whether better answers become better harvests is the open part.

Why now?

A chatbot can reply in a farmer’s own words, yet Rest of World quotes Catherine Nakalembe saying most AI models trained on European and US data are largely useless without local adaptation. Local questions and expert answers supply that training.

Who’s building it?

Digital Green builds FarmerChat with agronomists and veterinarians, while IFPRI and CIMMYT teams run the registered trials.

Digital GreenIFPRICIMMYTAgronomistsAIEP pilot teams
The full read
Why now, who is building it, where the market goes, and what stays unsettled.
Get weekly →
01 · The market
Who’s in it
Builders — Digital Green and pilot teams, with agronomists answering.
Buyers — smallholder farmers in Kenya, Nigeria, South Asia and Africa asking the chatbots questions.
Gatekeepers — trial teams at IFPRI and CIMMYT, who define what counts as proof.
What’s at stake
Whether local-language advice changes yields and household income, the outcomes that the Kenyan trial sets out to measure.
Who benefits from the data farmers supply to train models.
How well low-resource languages are covered by chatbots.
What’s unsettled
Whether high satisfaction turns into changed farming practice on the ground.
Whether voice delays and language gaps can be fixed.
Whether experts’ warning of data extraction from poor communities proves justified.
02 · How the demand matureda progression, not a timeline
Local data
Models built on Western data struggle locally; teams collect their own.
Local questions
Small models learn from over 120,000 farmer queries and answers.
Pilots
Five advisory pilots in Kenya and Bihar report an 800-farmer study with satisfaction of about NPS 60.
Trials
Registered trials in Kenya and Nigeria test effects on yields, income and fertiliser decisions.
03 · Where the market goesthe forks, not a prediction
→Satisfaction or yields: the trials may find farmers like the advice without changing practice, or find changes in yields and income.
→Language coverage may widen to low-resource languages, or stay limited by the labour of curating farm knowledge.
→Chatbots may communicate the uncertainty of fertiliser returns, as the Nigerian trial tests, or give single answers.
→Data may stay with local teams, or flow to firms training proprietary models.
The question underneath is evidence: advice that farmers rate well is not yet advice shown to change harvests, and the registered trials are the first test.
04 · The Bespoke Default deskadjacent markets, moving now
Hyperlocal AI
A language community owns its model.
Waikato team builds a Māori voice model owned by its speakers
IEEE Spectrum
Pace of One
Another personalised AI waiting on trial proof.
Randomised trial: AI tutor gains resemble Khan Academy practice without AI
NBER (working paper)
The Modelled Harvest
National statistics offices test AI crop estimates.
Philippine statistics office has begun piloting satellite and AI crop estimates
GMA News Online
05 · The Wire
01
Rest of World reports FarmerChat reaches over 1 million farmers.
Rest of World ↗
02
Pilots report satisfaction and three technical limits.
arXiv ↗
03
A 600-village Kenyan trial tests Farmer.Chat’s effect on yields.
AEA RCT Registry ↗
04
A 240-village Nigerian trial tests a Telegram advisory chatbot.
AEA RCT Registry ↗
The panel on this signal
They agreed on
–Advice for small farms is being built from local-language farmer questions and expert answers, led by Digital Green and pilots.
–Evidence of effect on yields and income is still limited.
–Latency, low-resource languages and curating farm knowledge are the pilots’ key technical challenges.
The behavioral economist disagreed on
–Satisfaction is not adoption. Advice changes harvests only if farmers act on it, so the trials measure practices and yields.
Know more about the method →
✦How this signal was built
01 · Panel
A Delphi panel of 4 AI analysts, each reasoning independently.
→
02 · Debate
They challenge and revise each other’s reads over two rounds.
→
03 · Evidence
Every claim is tied to a cited, linked source on The Wire.
→
04 · Synthesis
A moderator writes it into one read, disagreement and all.