Among the first robots most people encountered outside a factory was a floor scrubber working an empty supermarket aisle after closing. There is no arm, no attempt at conversation and nothing that resembles the robots of film. It maps the floor, drives a route, avoids what it did not expect and returns to charge. It is deliberately unimpressive, and that is the point.
The task suits a machine unusually well. Cleaning a large hard floor is repetitive, physically tiring and takes hours; it happens in an environment that barely changes between nights; and the standard is easy to define. It is also work that businesses struggle to staff reliably, because the hours are unsociable and the job is hard. Where the work is uniform and the surroundings are stable, a machine is a reasonable answer.
What these machines do not do is the rest of the job, and the difference is instructive. They cover the open floor and leave corners, edges, stairs, anything below a shelf and anything that needs lifting or judgement. A spillage that needs different treatment, a piece of packaging in the way, a mark that requires effort in one spot - these stay human. In practice the role does not disappear; it changes into supervising the machine, handling exceptions and doing the detailed work, with the large repetitive portion taken away.
The economics only work at a certain size. The machines are expensive to buy, are increasingly rented by the month, and pay back against hours saved - which means the calculation depends on how much open floor there is and how often it must be cleaned. A supermarket, a warehouse, an airport or a hospital corridor has enough of both. A small shop does not, and no amount of improvement in the technology changes that, because the constraint is the quantity of work rather than the capability of the machine.
The wider lesson generalises past cleaning. The tasks being automated first in physical work are the ones that are repetitive, happen in predictable surroundings and have an obvious measure of done. Work that varies constantly, happens somewhere different each time or requires deciding what the job even is remains stubbornly human, not because machines are not clever enough yet, but because those three properties are what make automation possible in the first place.