Agricultural data is “fragmented, distributed, heterogeneous, and incompatible.” That’s the verdict from a major Council for Agricultural Science and Technology report published barely a year ago, and it helps explain why AI has struggled to gain traction on farms. Other data-heavy industries, like healthcare or financial services, have established data standards, but agriculture has no universal framework for translating between the dozens of systems that generate field-level information.
This isn’t a new observation, but its persistence is noteworthy. While consumer tech and enterprise software largely solved their interoperability challenges years ago, agriculture still generates enormous volumes of information trapped in incompatible silos. Research institutions publish trial results in inconsistent formats, product manufacturers use proprietary naming systems, farmers record observations with local terminology and retailers track sales without connecting them to agronomic outcomes. The result is an industry sitting on massive amounts of information it can barely use.
“Agriculture doesn’t have a data problem—it has an intelligence problem,” notes Ron Baruchi, CEO of Agmatix, a company building domain-specific AI for the sector. “The data exists. What’s missing is infrastructure that understands what it means.”
According to a McKinsey report, implementing data integration, and connectivity in agriculture could add $500 billion in value to global GDP—a 7 to 9% improvement over current projections. But capturing that value requires solving a problem that general-purpose AI platforms have consistently struggled with.
WHY HORIZONTAL AI KEEPS FAILING IN FARMS
The appeal of applying large language models to agriculture is obvious: A farmer could describe what’s happening in their field and get instant advice on what to do about it, without hiring a consultant or having to wait for a lab. But agriculture’s complexity breaks the approach.
While an LLM trained on internet text might know that nitrogen helps plants grow, it can’t tell you that the right amount changes depending on the growth stage, the soil and what was planted in the same field the previous year. Similarly, computer vision can identify crop stress, but without contextual knowledge of weather, soil and product applications, that insight doesn’t mean much.
You can ask ChatGPT about nitrogen fertilization and get an answer that sounds authoritative. But when you dig into specifics—timing for your soil type, interactions with your previous crop, and product selection based on local availability—the recommendations fall apart.
