AI in farming: less tech, more human support

Yesterday it was blockchain; today it is artificial intelligence: every decade seems to bring its own miracle technology for transforming agriculture in the Global South. AI can genuinely add value in rural areas, from weather forecasting to agronomic advice. But a tool that is available is not a tool that is adopted, let alone one that is useful. A recent study by the Columbia Center on Sustainable Investment (CCSI) on Rwanda and Nigeria makes this point clearly, and its findings closely match what we see in our own smallholder programs. Hence a simple conviction: the usual proportions of digital agriculture projects need to be flipped, with a little technology and a lot of human support.

Agricultural AI: a new promise, a familiar script

About a decade ago, blockchain was set to revolutionize traceability, credit and insurance for smallholder farmers. FAO and ITU devoted a full publication to it, E-agriculture in Action: Blockchain for Agriculture, which already sought to demystify the technology and flag its implementation risks. A few years later, analysts were describing a form of “pilot-itis”: projects that worked at small scale but failed to reach larger populations, often because they had ignored key stakeholders at the design stage or applied the technology to problems it was ill-suited to solve.

AI is now on a comparable trajectory, with arguably broader potential. The CEIMIA review of AI in Sub-Saharan African agriculture maps concrete uses: crop monitoring, pest and disease detection, yield prediction, resource management and digital advisory. CCSI’s summary slide deck highlights applications already in use, from weather forecasting in Zambia to resource management in Nigeria and a farmer chatbot deployed across several countries.

Policy is following suit. According to CCSI, 14 Sub-Saharan African countries have a national AI strategy or policy, many of which name agriculture as a priority sector. Yet none has binding laws or regulations governing the use of AI in agriculture. Fundamental questions remain open: who owns a farmer’s data, who is liable when faulty advice leads to a lost harvest, and how are farmers protected from unfair outcomes?

This is not a case for rejecting AI, but for not expecting more of it than it can deliver. Columbia researchers put it plainly in a piece with an explicit title: AI’s promise requires innovation in governance, not technology alone. Their broader report on artificial intelligence and sustainability transitions likewise argues for weighing opportunities, risks and governance together.

Rwanda and Nigeria: what Columbia confirms from the field

The value of the CCSI study lies in its comparison of two very different models. Rwanda has opted for a state-led approach: its national AI policy sets a cross-sector vision, and its Strategic Plan for Agriculture Transformation (PSTA 5) explicitly embeds digitization and AI, with clear budgets and institutional responsibilities. Nigeria relies more on a dynamic startup ecosystem; its National Agricultural Technology and Innovation Policy (NATIP 2022–2027) emphasizes modernization but lacks a precise implementation plan, and responsibilities are spread across more institutions.

Despite these differences, both countries face the same five barriers:

• Incomplete governance, with no clear rules on data ownership, fairness or liability when AI tools fail.
• Uneven infrastructure: CCSI reports 4G coverage of around 95% in Rwanda, according to UNESCO, compared with around 53% in Nigeria, which limits access beyond major cities.
• Fragmented data across agencies, platforms and private players, without common formats or sharing standards.
• Limited, misdirected finance: the Gates Foundation-backed AI Scaling Hubs have received about $17.5 million in Rwanda and $7.5 million in Nigeria, but funding is concentrated on R&D and pilots; Rwanda’s AI policy puts the cost of full implementation at $76.5 million, of which only a small share has been mobilized.
• Last-mile realities: low digital literacy, limited trust, phones shared within households, and missing land titles or identity documents.

The last point is perhaps the most telling. In Rwanda, farmers aged 16 to 30 account for only 14% of the agricultural workforce, even though young people are the most digitally proficient. The country has therefore chosen to channel AI insights through extension agents, who turn outputs such as crop production forecasts into practical advice. In Nigeria, where around 70% of the population is under 30, direct farmer access to digital tools is more realistic, but the paid models offered by startups run up against smallholders’ limited ability to pay.

These findings closely mirror what we observe in the field. At the session Ksapa hosted on low-tech digital solutions for marginalized populations, the greatest hurdle identified was a severe lack of digital literacy, followed by a lack of suitable phones and reliable networks. Our lessons from Mobile World Congress 2025 pointed the same way: simply throwing technology at smallholders is bound to fail. Compatibility with basic and older phones, offline functionality and user testing with farmers themselves are what drive adoption.

Other research converges. An analysis on the London School of Economics’ Africa at LSE blog examines what it takes for digitalization to genuinely benefit African smallholders. CGIAR shows how AI can strengthen extension services rather than replace them. And ICTworks has long listed the reasons farmers do not use mobile agritech services, from shared phones to irrelevant content.

Flip the ratio: a little tech, a lot of human support

To put it schematically, many digital agriculture projects devote most of their resources, roughly 80%, to the tool itself and the rest to support. Our experience leads us to argue for exactly the opposite: a modest share for technology, including AI where it is justified, and the bulk of the effort on three pillars without which no tool works.

Content. A tool is only as good as what it delivers. Training modules must be co-developed with local partners, translated into farmers’ languages, illustrated and accessible to people with limited literacy. In our rubber program in Sumatra, the first training sessions focused on quick wins in cultivation practices, built on a thorough understanding of farmers’ core needs. The same principle applies to helping farmers meet certification requirements, where the app eases the administrative burden but does not replace the field officer guiding the farmer.

Infrastructure. This is CCSI’s second recommendation: build the foundations for AI before pushing AI itself. Rural connectivity, reliable electricity, device access and last-mile logistics often call for public investment. Project developers, for their part, must design tools that work with what already exists: an app that runs offline and on entry-level phones, like the SUTTI digital solution, reaches far more farmers than a sophisticated service requiring a recent smartphone and a stable connection.

Human support. This is the most underestimated and most decisive factor. Rwanda’s model confirms it: for older or less connected farmers, extension agents are the bridge between the algorithm and the plot. Our tips for successful digital tool adoption follow the same logic: training in small groups, simple access via SMS, and text-to-speech for those less comfortable with reading. A hybrid approach combining face-to-face and digital learning remains the condition for lasting adoption, as set out in our 10 keys to inclusive agricultural supply chain programs.

This shift in priorities applies to funders and buyers too. CCSI recommends making smallholder inclusion an explicit financing objective, using laws, licensing requirements, procurement rules and funding conditions that reward affordability, accessibility and farmer reach, not innovation alone. It also calls for combining public and private finance more deliberately: public and donor funding for infrastructure, extension services and early-stage deployment; private capital for scaling tools and services where viable markets exist.

AI has a real place in Global South agriculture, provided it is treated as an amplifier of a human system that already works, not as a substitute for one. The programs that succeed will not be those that pack in the most technology, but those that invest most patiently in content, infrastructure and the women and men who support farmers every day.

Image by Magnific – Free License