AI & the future of agriculture

Precision farming links observation ever more directly to the work in the field. Satellite images, drones and soil sensors map differences in growth, moisture and disease pressure within a single field. AI spots in the images where a crop is lagging, where a fungus is starting or where weeds are growing. The fertiliser spreader or the sprayer receives a task map and doses section by section. Cameras on the sprayer boom go further, recognising weeds plant by plant and spraying only there. The next step is a field managed plant by plant: each plant with its own record of growth, disease and treatment, kept by cameras on every machine that crosses the land.

In a smart barn, every animal wears a sensor, or cameras recognise the animals. They measure how much a cow eats, ruminates and walks, how a pig coughs and how hens spread out across the poultry house. AI compares those measurements with the animal's normal pattern. If its walking or eating changes, the farmer gets an alert, sometimes before noticing anything himself. That can point to lameness, mastitis or respiratory problems, but also to the right moment for insemination. Which conditions are picked up early differs from system to system. Changes that arise between two rounds of the barn become visible too.

Farm robots combine image recognition with physical work, more and more often with delicate crops and fruit. In the field, weeding robots recognise weeds and remove them mechanically or with a laser beam, without spraying; laser weeders are already on sale. In greenhouses, robots pick tomatoes, and robots for sweet peppers and cucumbers are being trialled on working nurseries. Image recognition decides which fruit is ripe and where the stem is. Today, such a robot usually does one task in one crop. The next step is robots that can learn different picking tasks, with grippers suited to the crop: the same machine picks strawberries in June and apples in September.

Gene editing makes targeted changes to genetic material possible; practical trials then show how well they work. The result is crops that cope better with drought or a fungus, and animals that do not fall ill from a particular virus. The FDA approved a gene edit that makes pigs resistant to the PRRS virus. AI helps breeders predict which genes determine a trait and what a change will do. Under the NGT regulation, certain gene-edited plants will be treated like conventionally bred plants after assessment; their seed will carry a label. The next step is a variety designed on a computer and growing in the field within a few years, instead of the ten to fifteen years that breeding often takes now.

For a long time, a grower set temperature, humidity, light, CO₂, water and nutrients by experience and with a climate computer. AI now combines measurements in the greenhouse, camera images of the plants, the weather forecast and the energy price, and works out which settings let the plant grow according to the cultivation strategy. A growing number of growers let such a system run the climate on its own, within limits they set. The grower sets the goals and keeps an eye on the crop. In international competitions, AI teams have already achieved higher greenhouse yields with less energy than experienced growers. The next step is a grower running ten greenhouses on three continents at once, with AI handling the day-to-day running of the crop.

Breeding with data increasingly connects DNA profiles with continuous measurements from the barn. AI looks for patterns in feed intake, movement and methane emissions, alongside the traits breeders already select for. That brings into view traits that used to be hard to measure at scale. Which traits carry most weight remains a choice for breeders and their organisations. The next step is more precise breeding for heat tolerance and the ability to recover, with data from thousands of barns at once. Selection will then give more and more weight to how well an animal keeps functioning as conditions change, now that heat stress counts for more with every warmer summer.

Emissions of ammonia, methane and nitrous oxide were long calculated with fixed averages per animal or per hectare, which show little of the differences between farms. Sensors in the barn now measure gas concentrations in the air continuously, and models combine those with the airflow to calculate the barn's emissions. AI links them to what happens on the farm: the feed, manure storage, ventilation, the timing of fertilising. Measuring farm by farm is now being trialled on experimental farms and in networks of working farms. A possible next step is permits and rewards based on a farm's own emissions, established with measurements and models rather than fixed averages.

How much comes off the land and what it will fetch is something a grower traditionally learns late in the season. AI combines satellite images of crop growth, weather data, soil maps and yields from earlier years, and predicts for each field how much will be harvested and at what quality. In the greenhouse, cameras count the fruit on the plant and forecast the harvest week by week. Cooperatives and traders add market data to estimate the price, and a grower can sell part of the harvest in advance. The next step is an agent that sells the harvest while it is still growing, batch by batch, with its origin and measured quality attached, to a buyer's agent on the other side of the world.

An agent is AI that carries out a series of actions itself, rather than just answering a question. Governments and suppliers are now trialling AI assistants that advise farmers, often by phone and in their own language. The next step is an agent that also acts. It connects data from barn and machinery systems with planning and execution: if milk yield drops, it adjusts the feeding plan; if the feed price falls below the farmer's threshold, it places an order; if the harvest forecast shifts, it rebooks the batch with the processor and arranges transport. Along the way, it deals with the agents of buyers and suppliers. The farmer sets the goals and the limits.

As AI and robots take over more work, jobs appear on and around the farm that barely exist today. The fleet manager of a robot swarm plans dozens of small machines across the fields of several farms, and steps in when a robot gets stuck. The remote grower runs several greenhouses from one place, sometimes on other continents. A carbon and nature assessor maps what a farm stores and restores, so that it can be paid for it. And there will be trainers who teach an AI model the judgement of an experienced farmer: what a cow looks like when she is slightly off colour, or when a field is really fit to drive on. The line between farmer, mechanic and analyst is blurring.

Leadership in agriculture means looking ahead and taking the first step. Leaders take the initiative on AI and new technology, even when not everything is settled. They sketch a picture of the future that everyone on the farm understands, from the worker in the barn to the young successor, and make it tangible with a small trial on their own farm. That is how they get family, members and staff to help build the farming of tomorrow. AI can model several scenarios side by side. Leaders decide where the business is heading and how to involve the people who work there.