Agriculture has a labour problem that no amount of wishing has solved: the seasonal workforce that weeds, thins and picks by hand is shrinking, ageing and getting more expensive, and in many regions the workers simply do not arrive in the numbers the crop needs. That shortage, more than any single advance in robotics, is what is putting autonomous machines into fields. When the alternative is a crop that cannot be tended or harvested at all, a robot that does part of the job imperfectly is competing against nothing, which is a forgiving benchmark.
The tasks that have taken hold are the narrow, repetitive ones where precision also saves money. Robots that identify and remove weeds plant by plant let a farm cut herbicide use sharply, because the machine treats the weed rather than spraying the whole field, and the same vision that finds a weed can thin a row of seedlings to the right spacing. Selective harvest — picking only the fruit that is ripe, and picking it gently — is harder and further behind, but progressing, because it is precisely the labour that is hardest to find. In each case the robot is doing a specific job well rather than replacing a farmhand wholesale.
The constraints are unlike those of a warehouse or a factory, and they are unforgiving. The robot works outdoors in dust, mud, glare and rain, on uneven ground, among plants that look different every week of the growing cycle. The window in which a given task must happen can be a matter of days, so a machine that is down for repair during that window has failed for the season regardless of how well it works the rest of the year. And the economics are seasonal too: the machine must earn its keep in the weeks it is useful, which shapes how these robots are sold, often as a service or a contracted operation timed to the crop rather than as a capital asset that sits idle for ten months.
The honest state of the field is that narrow tasks work and general autonomy does not, yet. A robot that weeds a specific crop reliably is a real, deployed product; a robot that manages a whole farm is not. But the direction is set by the labour arithmetic rather than by the technology's ambitions, and that arithmetic is not going to reverse. The machines that succeed are the ones aimed squarely at the job the missing workers were doing, built to survive a field, and priced to make sense across a season that is measured in weeks.