Agricultural mechanisation, which began with the replacement of human and animal muscle by steam and then by diesel power, has entered a new phase. Robotics, computer vision and precise satellite positioning are beginning to combine in machines that handle crops at a finer scale than a human worker typically could. Weeding, thinning, selective harvesting and targeted spraying are all operations that demand sub-centimetre accuracy, and they are moving one by one from prototype to commercial use on farms in Europe, North America and, more gradually, elsewhere.

The most developed area is weeding. Traditional herbicide spraying covers entire fields uniformly, even though most of the field contains no weeds at any given moment. Newer systems, often mounted on familiar tractor frames, use cameras to distinguish weeds from crop plants and apply herbicide only where weeds are detected. The reduction in chemical use can be substantial, up to ninety percent in some comparative trials. A Swiss company, ecoRobotix, developed a solar-powered weeding robot that patrols fields autonomously during daylight and targets weeds individually; similar machines have been deployed by several European manufacturers.

Thinning of vegetable crops has likewise been automated. In crops like lettuce, farmers historically planted seeds at dense spacing and then sent workers along the rows to remove surplus seedlings, leaving a single vigorous plant at each intended position. Vision systems now perform the same task, mounted on tractors moving at walking pace. Agricultural engineer Dr. Linh Tran, who has overseen field trials of a thinning system at the University of California, Davis, has observed that the machines match experienced workers in accuracy and substantially exceed inexperienced ones, a difference that matters because agricultural labour is increasingly difficult to recruit.

Selective harvesting is more demanding. A strawberry picker must identify ripe fruit, extract it without damage, and place it gently in a container, all under lighting that changes with the weather. Several prototypes now operate on commercial farms, though throughput is still limited compared with a skilled human picker. Dr. Tran has argued that the rate of improvement is the more important metric: current robotic systems pick a few fruit per second, but improvements of an order of magnitude are expected within a decade as computer vision and grasping technology continue to develop.

Beyond individual operations, several farms are experimenting with fully autonomous vehicle fleets. Instead of a single large tractor driven by one operator, the new model is a swarm of small, lightweight machines that can work collectively. Small machines compact the soil less, cause less damage if one fails, and can be deployed to avoid wet spots rather than forcing their way across a field. The economic case for these swarms is not yet fully established, but the technical case is increasingly plausible.

Critics have raised several concerns. Robotics may accelerate consolidation of farming into larger units, because the capital costs of the new systems are significant and difficult for small farms to recover. The loss of agricultural jobs, already substantial through earlier mechanisation, could continue. Data collection by robotic systems - what crops are in which field, what yields they produce, how management decisions are made - raises questions of who controls the data and for what purposes. Some farmers have expressed concern that large agricultural technology firms could extract data value without reciprocal benefit to the producer.

Environmental impacts are mixed. On the positive side, reduced herbicide use, less soil compaction and more precise water application all contribute to lower environmental footprint per unit of crop. On the negative side, the manufacture of complex electronic systems has its own environmental cost, and reliance on satellite navigation introduces dependencies on signals that can be interrupted.

Dr. Tran has argued that the transition to robotic agriculture will be uneven. Crops grown on large, flat fields will mechanise first, as they have in every previous agricultural revolution. Crops with more delicate handling requirements, or those grown on terraced or irregular land, will take longer and may be mechanised only partially. Regions with plentiful cheap labour will be slower to adopt robotics than regions with tight labour markets. The ultimate shape of agriculture in twenty years is likely to vary substantially among countries - more so, perhaps, than after the arrival of earlier mechanised technologies, because the new systems respond more flexibly to local conditions.