The evolution of digital technologies now makes it possible to apply artificial intelligence approaches in agriculture as well. The first point to clarify, however, is what we mean by intelligence.
For humans, intelligence is often associated with creativity. For machines, a more limited definition is enough: the ability to act rationally in specific contexts. An autonomous driving system, for example, does not need to be creative; it needs to move a vehicle safely and reliably, whether on a road or between vineyard rows.
In agriculture, AI applications can be developed with two broad goals. The first is to help perform more quickly and efficiently tasks that humans could already carry out on their own. The second, more ambitious one, is to solve problems we are not yet able to solve, such as inventing new agronomic methods capable of countering climate change. This second area belongs more to general AI and today remains closer to research than to widespread practical application.
Where AI is already useful
If we focus on the first type of application, practical uses are already visible. These include field robots, targeted distribution of phytosanitary products, efficient irrigation management, optimisation of winery processes and even the evaluation of marketing strategies by comparing them with practices already active elsewhere.
On a broader scale, such approaches extend beyond the single firm and can be used to analyse the suitability of territories for specific crops or to define which cultivations may be optimal in a context of climate change.
Among the most mature technologies, intelligent irrigation management is particularly relevant. By integrating soil humidity sensors, weather forecasts, mathematical models and automated distribution systems, farmers can receive precise suggestions on the optimal water input. The benefits go beyond water saving and can also positively affect the compositional characteristics of agricultural products, especially in viticulture.
Another fast-growing field is computer vision. Algorithms can extract agronomic or production parameters from images collected by satellites, drones, cameras or smartphones. Examples include mapping leaf volume and vigour, automatically identifying harmful insects through smart traps and estimating fruit size in real time during both growth and harvest.
Data and standards come first
These possibilities are exciting, but they depend on technological prerequisites. The first is the availability of annotated and accessible data. Making such data usable requires infrastructure for collection and organisation, accepted standards and the ability to combine information from different contexts. It also requires the collection of contextual agricultural and production information that helps explain the environment in which the data were generated.
This agronomic information is not secondary. It is the very basis on which AI models are trained. In that sense, agronomic and managerial data acquire enormous value. Farmers and associations need to become fully aware of that value when they negotiate with technology suppliers. Data must be shared to be exploited, but sharing has to be conscious and beneficial for all actors involved.
A system approach is essential
The integration of AI into the agricultural chain therefore requires a system-level approach. This must involve all actors in the chain, from policy-makers responsible for infrastructure and standards, to research bodies, trade associations and the producers themselves, who ultimately test tools and define priorities.
One of the biggest challenges is access to technology. It is unrealistic to expect agricultural professionals also to become experts in digitalisation and advanced technology. For that reason, entrepreneurial initiatives capable of translating AI into usable agricultural services must be encouraged. In many cases, especially with small and medium-sized farms, trade associations will need to mediate between technology providers and producers.
This immediately raises the issue of standardisation. AI models work better when trained on large quantities of data, but data sharing is not automatically advantageous for every business initiative. It is therefore necessary to decide which technologies are appropriate for sharing, which data should actually be shared and in what form.
Finally, investment in training is essential to create professionals able to mediate between artificial intelligence and agronomic intelligence. This is crucial to avoid AI solving problems that farmers do not actually have. The goal is not to create an agriculture without farmers, but to work with farmers to design the agriculture of tomorrow.
Contribution adapted from the 2025 Italian Food & Wine Tourism Report.
