One idea came up again and again: It’s not enough to showcase what AI can do—what matters is how it works in the real world.

AI tools can offer powerful capabilities, from detecting crop diseases to supporting farm-level decision-making. But they don’t operate in a vacuum. Their effectiveness depends on the systems around them—extension services, data infrastructure, financing, and public delivery mechanisms—that ultimately determine whether farmers can access and benefit from these tools.

The possibilities of AI within agriculture are real, however its greatest value will emerge when technological innovation is synergised with strong evidence, inclusive design, and responsive systems that ensure farmers—especially smallholders—can benefit from it.

Our rapid review of 51 studies on AI-enabled agricultural solutions across LMICs highlights both progress and important gaps. Much of the existing research focuses on predictive modelling and simulation-based studies—useful for demonstrating technical potential, but limited in what they tell us about real-world impact. Far fewer studies look at outcomes that matter most to farmers and policymakers, such as income, food security, sustained adoption or who actually benefits.

In other words, we are getting better at understanding what AI can do, but we still know far less about what it actually changes.

There are, of course, encouraging examples. AI-powered crop diagnostics are helping farmers identify diseases earlier and reduce losses. Predictive tools are improving advisory services, supporting decisions on planting, irrigation and fertiliser use. These innovations point to the real potential of AI to strengthen agricultural decision-making.

But moving from potential to impact—and from pilots to scale—remains a challenge. Many promising tools stall because they are not embedded in the systems farmers already rely on, such as extension services, insurance schemes or supply chains. At the same time, persistent barriers like limited connectivity and digital literacy continue to shape who can access and use these technologies.

Across the discussions, three challenges surfaced repeatedly: fragmented agricultural data ecosystems, gaps in digital infrastructure, and institutional systems that struggle to keep pace with rapid technological change. But just as important are the questions of equity and context that sit beneath these challenges. Without careful design, digital technologies risk benefiting farmers who already have access to infrastructure while leaving others further behind.