TL;DR
Construction automation has only ever worked on two kinds of tasks — those you can move into a factory and those repetitive enough to be done factory-style on site — and nearly everything else has failed for 170 years.
Building a single-family house in the US spends close to half its cost on labour. Building a car spends six to eight per cent. So it is no surprise that people have been trying to put machines on the job since 1853, when Isaac Hussey patented a plastering machine.
The author turned AI loose on more than a century of construction trade journals to see what had been tried and what survived. The conclusion is blunt: only two kinds of tasks made it.
Move it into a factory and it works
The first kind is work that can leave the site altogether: steel fabrication, precast concrete, windows, roof trusses. These are mass-produced components, no different in principle from any other manufactured good, so they automate.
Concrete took the same road. In the 19th century it was mixed on site with shovels; production moved off-site, and by the Hoover Dam the batching plant was fully automated. Windows and trusses followed the same path from site-built to factory-built.
Make it factory-like and it works
The second kind stays on site but is monotonous enough that a machine can creep along a simple path, doing the same motion again and again.
Concrete and asphalt are the stars here. Concrete is a thick viscous fluid: squeeze it into a mould and it holds the shape, which is why slipforming machines can extrude kerbs and pavements in a continuous ribbon.
Roads, railways and tunnels follow the same logic — a machine moving along a line, repeating one operation. Rail maintenance, replacing sleepers and tamping ballast, is startlingly automated.
Why bricklaying keeps failing
Automatic bricklaying is the classic failure. The task looks ideal: backbreaking and wildly repetitive. Machines have been tried since the early 20th century and historically none of them worked.
The catch is that it is not quite repetitive enough. Spreading mortar and setting a brick so that both the brick and the wall stay level is hard for a uniform mechanical motion; bricks come out slightly askew and need individual nudging.
Walls are not uniform either. Door openings, windows, corners, joints, short runs. Every one of those stops the machine and needs a human to reset it, and resetting every few minutes is worse than just laying the bricks by hand.
Japan is the cautionary tale. From the late 1970s it built more than a hundred kinds of construction robot, then wrapped whole towers in automated 'skyscraper factories'. Setup costs were enormous, small buildings could not justify them, and big ones had to be designed around the robots from the start.
This time may really be different
The old failures came down to information processing. A machine did not know where it was, what was in front of it, or whether to adjust. It just repeated one motion forever.
Modern AI is good at exactly that. Feed a model enough training data and it handles situations it has never seen. Startups founded after 2022 have taken roughly half of all venture funding in construction robotics.
Whether the future is a fleet of general-purpose humanoids or a zoo of single-task machines, the author will not guess. But he is confident of one thing: the future of construction automation will not look like its past.
Curated from high-quality sources, with concise summaries and key takeaways.
