AI Is Agriculture's Next Big Opportunity. But Which Decisions Has It Actually Changed?
Aidan Connolly, President, AgriTech Capital, is a food/feed/farm futurologist and author of the book The Future of Agriculture.
gettyIn a new survey of nearly 2,000 agri-food professionals, I asked where they see the greatest opportunities for AI and how they are using it today. The respondents were primarily agribusiness executives, consultants, technology providers and researchers, so the findings reflect industry perspectives.
Overall, their answers were unsurprising: 67% identified better decision-making as a major opportunity, and 65% pointed to productivity. However, the picture became more interesting when we asked about specific applications. Just 5% identified autonomous equipment and robotics as potentially the most significant application, while 31% selected decision support and recommendations.
These findings raise an important question: How does agriculture move from recognizing AI’s potential to using it in ways that genuinely improve business performance? The answer may have less to do with choosing the right technology than with identifying the decisions worth improving.
Twenty-nine percent of the respondents say that they have used AI features built into software they already owned, compared with 9% who had used a tool developed specifically for agriculture. That number suggests two things: First, it may undercount actual usage, as AI features are now tightly bundled into office software. More importantly, it indicates an opportunity for increased use of AI tools in both commercial and production decision making, and that the biggest AI opportunity for agribusiness right now may be in purchasing, inventory, scheduling or other commercial decisions.
McKinsey’s 2026 global survey, The State of AI: On the Road to ROI, found that 80% said AI had improved their individual productivity, while 37% attributed at least some EBIT impact to AI use. Although only 6% were classified as AI “high performers,” nearly 75% of those had fundamentally redesigned workflows because of AI (versus 25% of the other respondents).
Using AI to perform an existing task faster is different from redesigning how work gets done, but the latter is much more closely associated with meaningful returns. It also gets harder when the recommendation affects a biological or operational outcome in food and farming.
Generating recommendations about feed, treatment or nitrogen timing is very different from creating a document. The recommendation has to be specific enough to act on, providing the right information, at the right time.
Over the past two months, I’ve moderated conference panels in China, the U.K. and the U.S. At each, I’ve asked a simple question: Has AI actually changed a task or a decision in your business? Very few people have been prepared to say yes.
That gap between AI’s potential and actual impact matters particularly in agriculture. Saving time drafting an email is useful. Changing a nutrition, treatment or crop-management decision is a different test: Someone must act on the recommendation and live with the result. The University of Wisconsin-Madison asked 44 independent soybean experts to evaluate AI-generated management plans and found that the recommendations missed important regional differences. The conclusion was that agricultural AI needs to move toward tailored, field-specific advice before farmers can just trust it.
Adoption will partly depend on the user trusting recommendations enough to act on them. McKinsey’s Global Farmer Insights 2026 survey of 5,500 farmers across 10 countries found that just 17% are using AI for farm-related tasks, and only 6% cited generative AI or AI search as a trusted source for decision-making. Our respondents raised similar concerns: Integration with existing systems and accuracy were cited as serious concerns by 32% of respondents. Only lack of knowledge or training was seen as a more serious obstacle, at 52%.
McKinsey’s The State of AI in 2025 survey found that inaccuracy was the most common AI-related risk, and that high-performing organizations were more likely to have defined when model outputs need human validation. Our survey did not ask who should have the final say, but my view is simple: If an AI recommendation can affect an animal, a crop or the bottom line, somebody still needs to own the decision. Trust shouldn’t depend on how confidently AI delivers a recommendation. It should depend on how consistently those recommendations perform against the alternatives, and on knowing when human judgment is still required. Its track record should determine how much responsibility it earns.
So where should an agribusiness begin? With the decisions, not the software. I’ve argued before that AI pilots should address specific business challenges, and the same principle applies here. Which recurring decisions have the greatest impact on margins: feed rations, treatment protocols, input purchasing, production scheduling? Who makes those decisions today, what information do they use and what would a wrong call cost? That last question really matters.
Start with a decision whose outcome can be evaluated relatively quickly, rather than one you can only judge at harvest. Choose one where the available information is reasonably good and the cost of error is manageable. Test the AI recommendation against the existing approach, identify who has authority to accept or override it and keep a record of what was recommended, what was done and what happened.
Then evaluate the results. Was the recommendation accurate? Was it used? Did it improve the outcome compared with the existing process? Building confidence will generally take repeated testing, but this is how experimentation becomes evidence.
Our survey respondents identified lack of knowledge and training as the biggest barrier to adoption. Their open-ended responses reinforced the need for practical training, better integration, confidence in accuracy and clearer evidence of value.
Before you ask which AI tool your business needs, ask which decision you most want to improve. What does it need to know, and how will you tell a season from now whether the recommendation was right? If you cannot answer those questions about the decision, buying another AI tool will not solve the problem.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
