Contractors Aren't Buying Robots

Last week Jake Hall posted a list of eighty-odd robot companies in construction and asked who he had missed. The best answer under it wasn’t a company. It came from Cliff H., who builds circuit boards. If a robot saves a hundred hours of labor and creates fifty hours of downtime, troubleshooting and workarounds, he asked, what did it actually deliver? His answer: contractors aren’t buying robots. They’re buying predictable results on the jobsite.

He is right. And he stopped one step short of the thing that matters.

Dependability is not a property of the machine. It is a property of the crew formed around it.

Start with the size of the thing. Americans spent construction dollars at an annual rate of about $2.2 trillion in August. Forecasts for the global construction robot market by 2030 run from $164 million to about $3.7 billion, depending on whose report you buy. Even the generous number is about a sixth of one percent of a single year of American construction spending.

The robots are a rounding error. Adoption has barely begun.

That is the good news, because it means the habits are not set yet. And the pressure to set them is real. Construction job openings swung from a record low of 202,000 in February to more than 300,000 in June, and the economists at ABC point to immigration enforcement and retiring trades as the reason. This is not a labor-cost problem that a robot solves by being cheaper than a man. It is a labor-availability problem. The robot is being asked to stand in for people who are not coming.

So who stands next to the robot?

I learned the answer on a glass line at Lansing Delta Township, where we built the Traverse, the Acadia and the Enclave. Robots laid the urethane bead around each windshield and backlite. People set the glass, by suction cup on a winch. A clean division of labor: the machine did the repeatable thing, the human did the placing.

One shift the vision camera misread the option code. It called for a rear glass with an antenna the build didn’t need, and the robot dutifully beaded it. No machine caught it. The quality inspector at the end of the line did. That inspector worked for Karen, my quality team leader, and Karen got Abdelhak Radjai on the radio. Abdelhak ran that line with me. He walked it back, found four cars with the wrong back glass, and rallied the team leaders.

We stopped the line several times that shift. They cut the glass out and cleaned the urethane off the body. They ran new glass through the bead robot. There was no deck and no winch for this, so two people set each glass by hand at line height without smearing the headliner or the inner quarter trim. Four cars rebuilt, a few minutes lost, nobody hurt.

Write that day up in Cliff’s arithmetic. The robot generated the defect. The formed crew absorbed it, contained it, and kept the plant’s promise to the next station. If you had measured that robot’s “dependability” on that line, you would have been measuring the crew and crediting the machine.

I have stood beside a lot of robots in 36 years. I never once saw one deliver its rated uptime on its own. The uptime came from someone who knew what the fault code meant, who staged the material before the machine starved, who noticed the bead looked wrong before the camera did. The spec sheet promises the output. The people keep the promise.

Two companies on Jake’s list seem to have understood this from the start.

Canvas builds a robot that tapes, muds and sands drywall. Before it went to customers, it went to the painters’ union. Development and testing happened inside the International Union of Painters and Allied Trades’ own training facility. Canvas became a union signatory, and it doesn’t sell the machine as a gadget. It trains and hires union finishers and works as a subcontractor. Construction Dive reported the system cutting a seven-day finishing job to two. Notice what was actually delivered: not a robot, but a formed finisher with a robot in his hands.

Construction Robotics built SAM, a semi-automated mason. SAM does not replace the mason. One mason runs it, a tender keeps it fed with brick and mortar, and another mason sets the wall ties and lays the corners. The company claims one mason plus SAM can match four or more masons. Then one of its own people said the quiet part plainly: the robot lays 350 to 400 bricks an hour only if it has bricks, has mortar, and the crew is prepared. The machine forced the crew to plan. The planning is where the productivity lived.

Those are vendor numbers, and I would want to see them on a contractor’s own books before I bet a business on them. But look at the design choice, not the brochure. Both companies built the machine to multiply a trade, not to replace one. Both made the trained tradesperson the unit of sale.

What does it look like when someone goes all the way? Construction has no clean case yet. The nearest one is next door, in mining.

Fortescue put Caterpillar’s autonomous haul trucks into commercial service in the Pilbara in 2012 and converted its whole haul fleet, 183 trucks, by 2020. The company reports a 30 percent productivity gain over driver-operated fleets, and its CEO tied that technology push to staying at the low end of the global iron ore cost curve. But read the fine print. Fortescue trained more than 3,000 people to work with the autonomous fleet, 200 of them as mine controllers. The drivers did not simply vanish. A new trade was formed to run the system.

Then the counter-example. Katerra raised about $2 billion from SoftBank to industrialize construction end to end, factories and all, and filed for bankruptcy in 2021. It tried to engineer the trades out of the building. The buildings refused.

So does full adoption solve labor, growth and margin? Where it has worked, the gains are real: up to a third more output from the same people. Nowhere did it work by removin