Two widely shared clips claim that machines have replaced workers on city streets and building sites. The events are real. The story told about them is not.
The short video is persuasive. A robot dog trots down a San Francisco pavement while a caption says a startup pays twenty dollars an hour for such machines to roam the city, gathering information by any means. A second clip shows machines raising the brick walls of a house in the Netherlands. Its narrator says an American firm built an apartment block with “human robots”, that people only laid the slabs, and that fifteen new orders followed. Both clips end on the same thought: the age of human labour is closing.
Each clip grows from a real event. Each also changes that event in the same direction, by removing the people who made it happen. Checking them is a useful exercise, because the same pattern runs through almost every robot video now in circulation.
The dog on a lead
The startup exists. Dimensional, based in San Francisco’s Dogpatch district, did advertise work at $20 an hour, in shifts of four to six hours. But the money went to people, not to robots. The job was to walk a four-legged Unitree robot through the streets, steering it with a handheld remote or a phone app. No robotics experience was needed. Applicants had to be willing to carry the machine, which weighs about 33 pounds, back to the office if it overheated and stopped.
The reason is the opposite of the one the video suggests. The company is building dimOS, open-source software meant to let AI agents control many kinds of robot through plain-language commands. To get there it needs recordings of competent navigation, which the founder, Stash Pomichter, has called “ground truth”. The dog’s cameras and sensors logged each human-guided walk so that the routes could be compared with those an AI agent would choose. Pomichter told the New York Times, which reported the story in late October 2026, that agents are still very poor at controlling hardware. The listing has since been deleted.
So the clip’s central image is true and its meaning is reversed. The dog was not roaming. It was being taught, by a person on an hourly wage.
The mason without legs
The second video distorts more. Monumental is not American; it is based in Amsterdam. Its machines are not humanoid. The lead robot, called Pisa, has been described by trade reporters as two small tower cranes with two arms, one placing bricks and the other laying mortar. Two support robots, Petra and Panama, keep it supplied. Software named Atrium reads the architect’s drawings and plans each movement.
What the machines achieved is still notable. On a site in Almere they laid every brick of a house, including curved walls, mixed brick types and lettering set into the brickwork. That is a first for the company, whose robots have now worked on more than a hundred homes in the Netherlands and the United Kingdom.
But it was a house, not an apartment block, and the robots built only its brick walls. Builders did everything else, from the foundations to the roof. The chief executive, Salar al Khafaji, has said that fully autonomous homebuilding is decades away.
The “fifteen orders” are a misreading too. Al Khafaji said that another fifteen houses were due in the coming weeks, which describes a work schedule, not customers won over by a demonstration. The company does not sell robots at all. It works as a subcontractor and is paid for finished walls, so that builders need not buy or run the machines. Its next project is a neighbourhood of 112 homes and a hotel in Groningen, for families whose houses were damaged by earthquakes, and it plans its first pilots in the United States.
The needle and the cloth
The same family of videos includes footage of humanoid robots learning to sew from garment workers. Here the record is thinner. Jack Technology, a large Chinese maker of industrial sewing machines, announced in June 2026 that it would use Siemens software to bring AI and humanoid robots into sewing workshops. Trade publications reported in July that it had ordered 2,000 humanoids for apparel work.
The company’s own executive supplied the caveat. Lou Lingtong said the robots can perform individual sewing operations but cannot yet run a complete production line. Early pilot results came from a single product and a single fabric. Cloth is soft, stretchy and inconsistent, which is why sewing has resisted automation for so long. If the approach does scale, trade analysts note, the effect would be felt most in countries such as Bangladesh, Cambodia and Ethiopia, whose garment industries rest on low wages.
How the distortion works
Set side by side, the three cases show a method. First, the robot is made more human: a twin-armed crane becomes a “human robot”. Second, the humans vanish: the dog’s handler, the builders, the line workers who still finish the shirt. Third, a plan becomes an accomplishment: scheduled houses become orders, and a pilot becomes a takeover. Finally a famous name is attached, so that a forecast about the distant future seems to describe the present.
None of this means the machines are trivial. A robot crew that lays every brick of a house is a serious advance in a trade short of workers. A company that pays people to walk robot dogs is betting that the dogs will soon walk themselves. The direction of travel is not in doubt.
What is in doubt is the speed, and the claim that people are already unnecessary. In each verified case, humans steered the machine, supervised it, supplied it or completed the job around it.
A viewer who wants to judge the next clip can ask three questions. Who is holding the remote? What did the machine not do? And who is telling me this, and what are they selling?
Those questions will not make the videos less exciting. They will make our reading of them more accurate. That matters, because choices about training, careers and public policy are already being shaped by what people believe machines can do.


