Agriculture Has Adopted AI. So Why Isn't It Transforming the Industry?

Over the last few months, we surveyed more than 2,000 senior agri-food professionals about their use of AI. I expected to find an industry interested in AI, but still cautious about adopting it.

That isn't what we found.

Around 60% of respondents are already using AI every day, and 84% use it at least weekly. Three-quarters report a positive return from it. Around 90% expect to increase their use over the next year. For a technology this young, those are extraordinary numbers. In two decades of watching technology cycles in this industry, I can't recall sentiment being this one-directional.

But there is a catch: AI has reached the office, but not the field, and not systemically. If you look behind the headline number at what AI tools people are using a very different picture emerges:

• Writing and communications: 83%.

• Research and information gathering: 81%.

• Data analysis and reporting: 59%.

That’s fine, but it’s basically just an improved version of familiar tools.

Now look at some key activities, right at the heart of agriculture:

• Crop management: 3%.

• Agronomy recommendations: 6%.

• Livestock management: 10%.

• Precision agriculture: 11%.

Fewer than one in ten respondents have used an AI tool designed specifically for agriculture, so let’s examine that finding a little more closely. The responses suggest five obstacles:

1. An AI capability gap

More than half of the respondents cited a lack of knowledge or training, whereas just 13% cited cost. That makes sense when you consider how AI arrived.

Previous waves of agricultural technology tended to reach the industry through established channels: extension services, cooperatives, suppliers, consultants and other known advisers. Generative AI, on the other hand, arrived as an app.

Typically, adoption of a new technology starts with users finding the features and tools that are familiar, so when people opened ChatGPT, Claude, Copilot or Gemini and started experimenting they found the low-hanging fruit: search, but better; word processing but better, reports but better. So far, so good.

However, the institutional capability hasn't caught up, so people are moving very slowly on the advanced features that offer the opportunity for real change.  Features such as automation, advanced analytics, collaborative workflows require both a change of work flows and learning new systems. AI tools have the hook- do existing jobs better and faster, but now they need to provide a scaffold for users to map their business processes onto new systems. The comments bring this to life:

• “You don't know what you don't know.”

• “Finding a person to help with intentional AI use for our people.”

• But also: “Once you help people to use AI correctly, they never look back!”

In short, the problem is not persuading people to use AI, it is providing mechanisms to help people figure out how to use AI tools well in their organization.

2. The foundations aren't always ready

The next barriers derive from the first: respondents cite the challenge of integrating AI tools with existing systems (32%) and problems with data quality or availability (20%). Implementation is where theoretical benefits hit reality, and figuring out how to break it down into manageable modules is a real challenge. The quality of data available is always an issue in technology adoption, and agricultural data is particularly difficult. In addition to the usual challenges of incomplete and inaccurate data that plague most systems, agricultural data is typically siloed across machines, software platforms, suppliers, farms and seasons that are not designed to talk to anything else.

One respondent gave perhaps the best warning in the survey: adding AI to broken processes and poor-quality data simply creates more noise and confusion.

There is a temptation with any exciting new technology to start with the technology. In many businesses the better starting point will be much more mundane: What information do we have? Where is it? Can we trust it? And which business process are we trying to improve?

3. The AI industry hasn't really come to agriculture yet

This was a particularly interesting finding. The respondents use an average of three or more AI platforms, but overwhelmingly they are the same general-purpose tools being used everywhere else. Fewer than one in ten have used an AI product developed specifically for agriculture.

Part of the explanation may be structural: agri-food has a lot of small and mid-sized businesses, often geographically dispersed and operating on relatively thin margins. They aren't obvious targets for the huge enterprise AI implementations being sold into banking, pharmaceuticals and other sectors. But that isn’t a full explanation: there are some very big and centralized sectors within the agri-food category.

The challenge as the Chairman of one of the largest integrated pig producers said is ‘the largest consulting groups and mega AI players just don’t understand agriculture, our language, our challenges and our goals.’ The disjointed nature of agri-data, and the continued reliance on human input, creates opportunities for AI to arrive at the wrong conclusions. It has also created the opportunity for smaller specialized players to fill the agri-food space (e.g. Swarm Engineering) left open by the AI superpowers.

One processor noted that “I wish our phone was ringing off the hook from technology companies pursuing our industry. Unfortunately, the phone has not rung once.” That's worth paying attention to the assumption is that agriculture is just slow to adopt technology. Perhaps technology companies also need to ask how hard they have tried to serve agriculture.

4. In agriculture, getting it wrong matters

Accuracy was cited as a barrier by 32% of respondents, with 25% mentioning privacy and security. This is rational, not resistance to technology.  If an AI tool hallucinates a paragraph in the first draft of a memo, someone corrects it and you have lost a few minutes, but a bad agronomic recommendation can cost a crop. A bad livestock decision can have serious consequences for both the animals and the farmer. A mistake in a food-production environment can have critical safety implications.

The tolerance for error therefore decreases as the integration of AI into the operation increases. As one respondent noted: “You only need to suffer an AI hallucination once to become doubtful.”

5. Personal productivity isn't the same as transformational advantage

This may be the biggest issue. Nearly half of our respondents believe that AI can be transformational and can create major competitive advantage, but most of the other half see it principally as an efficiency tool.

The difference between those outcomes may be less about technology than about what companies do with it. Executives and their employees are using AI productivity tools, saving time, producing first drafts faster, researching more efficiently and getting useful analytical support. Those are real gains.

But giving everyone an AI license isn't an AI strategy, and without deeper integration into the organization, AI tools are limited in their effect.  If ten employees individually save two hours a week, you have an efficiency gain. If the company uses AI to rethink how a process works, how decisions are made, how information moves or how customers are served, there is the potential to create exponential change. Efficiency is what you get from adoption. Advantage is what you get from redesign.

Supply chain optimization is often recognized as being the place to see the quickest wins.   If a commodity crop is being moved significant distances to be processed into feed or meet a specific customer demand, driving past other possible processers with spare capacity, could the same be achieved with significant savings.   AI can process these decisions better, faster, and I’ve already seen examples resulting in millions of dollars of savings.

So where do we go from here?

Among a group of senior, relatively engaged agri-food professionals, individual AI adoption is already extremely high, yet penetration into core agricultural applications remains extremely low. This survey gives us a pretty good indication of what the industry wants.

When we asked where AI could create the greatest value, the leading answers were better decision-making (67%) and improved productivity (65%). Decision support and recommendations were also the clear winner when we asked which applications would matter most over the next five years.

For all the excitement about autonomous equipment and futuristic farms, the people running and advising agricultural businesses are asking for something much more immediate: help us make better decisions, using the information we already have. That is a sensible place to start:

• Build people's capability.

• Prepare the ground, in particular getting the data and processes into reasonable shape.

• Put sensible governance around AI tools in place

• Begin with tools and modules that support people making decisions before trying to remove them from the process.

• But then change the question from “Where can we use AI?” to “If we had new or better tools and information, would we still run the business this way? What would we do differently?”

Our survey suggests that people in agriculture aren’t waiting to be persuaded about the value of AI, they want more accessible options. The opportunity now is to turn that enthusiasm into something that changes not just how efficiently people work, but what agricultural businesses can do.

Thanks to my co-researcher John Power, LSC International, Abi Santos for coordinating the survey and data collation, and Kate Phillips Connolly for editing this blog.