Six questions every South African farmer should ask before signing for AI
The agri-tech salesman at the farm gate is selling software against a seasonal payback, and the contract decides whether the farmer or the vendor carries the loss.
Sign up for my Substack daily AI newsletter here.
Follow me on LinkedIn: https://www.linkedin.com/in/johanosteyn/
There is a strange split in South African agriculture today. On one page, consultancy reports celebrate farmers as the fastest adopters of artificial intelligence of any professional group on earth. On another, surveys of actual grain producers show more than half saying the technology is simply too expensive, and a surprising number still writing their records on paper. Both can be true at the same time, and both are. The reason sits at the farm gate, where a widening gulf has opened between what agri-tech is sold as and what the product is actually doing when the contract takes effect. Before any South African farmer signs the next three-year subscription, six questions deserve a written answer, and the answers are rarely in the brochure.
CONTEXT AND BACKGROUND
South Africa has quietly been one of the more sophisticated precision-agriculture markets in the Southern Hemisphere for over a decade. Yield monitors, variable-rate fertiliser prescriptions and satellite imagery for crop-stress mapping are already established, and the discipline underneath them is real. What has changed in the past two years is the retail layer on top. Software platforms, cloud dashboards, subscription tools and generative AI assistants are being added to the shopping list, often by vendors with no track record in agriculture. A national survey published this August by SA Grain found that among eighty-three South African grain farmers, more than half of non-adopters cited cost or the absence of a business case as the main barrier, and nearly a third still keep their records mainly on paper. The gap between the pitch deck and the field is where the next contract is being signed.
INSIGHT AND ANALYSIS
Three risks hide inside the AI label. The first is data ownership. When a farmer cancels a platform in three years, do the yield records, spray logs and soil tests leave in a format another programme can open, or do they stay on the vendor’s server? American farmers have already fought that battle. In April 2026, John Deere settled a right-to-repair lawsuit for roughly one point eight billion rand, with farm data ownership at the centre of the dispute. The same subscription language travels with the tractor to the dealership in the Free State.
The second risk is vendor mortality. AppHarvest, once a flagship of American indoor farming, filed for Chapter 11 in 2023, and Bowery Farming closed in November 2024, leaving customers with hardware disconnected from any functioning service. A contract that assumes the vendor will outlast it is making an assumption the market itself does not support.
The third risk sits in the recommendation layer. AI can be confidently wrong: a spray recommendation on bad soil data, a feed ratio against a mislabelled sample, an irrigation schedule on drifted sensors. When the loss arrives, the contract blames the farmer, not the model. This is where cognitive governance belongs on the farm, not as a slogan but as a working rule: that the automated recommendation is a draft and the farmer is the signing authority.
IMPLICATIONS
Behind these three risks sits a fourth. Farm management platforms routinely route data to servers outside the country, often in Virginia, Dublin or Frankfurt. If that stream includes worker biometrics or payroll figures, the farmer is a responsible party under the Protection of Personal Information Act (POPIA), and the Information Regulator will treat the farmer as the point of accountability. The European Union now operates an industry code on how agricultural data sharing should be contracted. South African agri-tech contracts rarely mention anything equivalent, which puts the burden of thinking it through on the buyer.
The practical consequence is that the three-year cost of an AI product is almost never the invoice at signing. Subscription escalators, connectivity, hardware refresh and support turn a competitive quote into a liability by year three. A disciplined farmer models those costs before the pen moves, and asks for written answers to the data, vendor and liability questions in the same meeting.
CLOSING TAKEAWAY
Agriculture is one of the few sectors where the person at the signing desk also bears the operational risk of what gets signed. There is no corporate liability shield between the farmer and the weather, the input cost, the market price or the AI recommendation. That is why procurement discipline on the farm has always been among the toughest in business. The six questions deserve to be put to every vendor, and the answers deserve to arrive in writing. A good vendor will welcome them. A vendor who hedges has already answered.
Author Bio
Johan Steyn is a prominent AI thought leader, speaker, and author with a deep understanding of artificial intelligence’s impact on business and society. He is passionate about ethical AI development and its role in shaping a better future. Find out more about Johan’s work at https://www.aiforbusiness.net
Comments