The electric motor took decades to create the kind of factory we now take for granted. To see why, you have to stand inside the thing it replaced.
For a century and a half, from the late 1700s into the 1920s, much of factory machinery was driven from a central source. In many early mills, that source was a water wheel. Later, it was an enormous iron steam engine, sometimes as large as a small house, with workers shoveling coal into its boilers throughout the day. The source changed, but the way power moved through the factory did not. From that central drive, an iron shaft ran the length of the ceiling. Every machine on the floor, every lathe, every drill, every press, was connected to that overhead shaft by its own leather belt running down from a pulley above. That belt was the only way power reached it.
The largest mills ran several hundred machines this way, stacked across floors and arranged by how close they sat to the engine, all driven by that one shaft. Together, the engine, shafts, belts, and machinery filled the floor with a deafening roar, while someone walked the line oiling the bearings by hand as oil worked loose and fell from overhead. If the shaft turned, your machine could turn. If it stopped, every machine stopped with it, in the same instant, ready or not. Hundreds of machines, one engine.
In the 1880s, the electric motor arrived, and the obvious thing to do was the thing nearly everyone did.
They pulled out the steam engine and put a large electric motor in its place. Everything else stayed. The same shaft ran the ceiling, the same belts came down, and the same building stood exactly as it was. The motor turned the line the way the engine had, and the floor kept its single will.
And this was not a mistake. Swapping the engine for a motor was the sensible first move. It proved that the new power source worked, ran cleaner, and demanded less attention than tending a boiler. You would be right to do it.
But that first comparison could reveal only so much. A new source of power, poured into the old shape, could do only what the old shape allowed. It could prove that electric power was capable of replacing steam. It could not yet show what might follow once the factory no longer had to be built around a single source of power.
The real change took four decades to arrive, and it did not look like a better engine. Once electric motors became compact enough to mount on individual machines and affordable enough to install across the factory floor, the line shaft could disappear. No shaft overhead. No belts dropping from the ceiling. No central motor driving the whole floor.
Before, every machine depended on the same moving shaft. Pulleys, clutches, and belt shifters could adapt that power and let a single machine sit idle, but the shaft itself had to keep turning whenever anyone was working, and a large share of the power was lost to friction in the shafts and belts before it ever reached the work. With its own motor, each machine could be matched to the job, with the right size, power, speed, and gearing. It could run at the speed the work required, draw no power while stopped, and operate on its own when the rest of the floor was dark.
That changed more than how the machines were powered. Buildings no longer had to be designed around the path of the shaft. Machines could be arranged around the flow of the work instead of sitting in lines parallel to the power. What had been one rigid system became a factory that could be rethought and rearranged as the work changed.
The greatest value was not in the motor itself. It was in the constraints the motor let you drop, and in what you could build once the old rules no longer applied.
When I sit with teams looking at Physical AI, the room usually holds several reasonable views at once. Engineers see a system that has to integrate. Operators see something that has to earn trust in an environment where things already work. Product teams see a capability that has to create repeatable value for customers. The people responsible for the budget see an investment that has to produce a return.
A focused use case gives everyone a practical place to begin: find a job the new system can take on, measure it against what already exists, and demonstrate a meaningful improvement. That proof matters. It shows that the technology can work under real conditions, builds confidence, and creates the evidence needed to take the next step.
But once the technology has proved itself, the question widens. Which parts of the system were designed around limitations that no longer have to apply?
Start somewhere more fundamental than the technology and ask what changes when intelligence can operate directly in the physical world. For most of industrial history, understanding what was happening across factories, field operations, infrastructure, and public spaces depended on people or central systems gathering the signals and making sense of them.
Physical AI adds another possibility: systems that can interpret what is happening where it happens, whether on a production line, across a remote operation, throughout a city, or in the moments leading up to a safety incident. That gives people greater awareness, faster understanding, and more room to focus their attention and act where human judgment matters most.
First, what if the interface could help people act sooner instead of simply giving them more to watch?
This is a common conversation in my work with customers. In utilities and logistics organizations, hundreds of sites feed data into a central operations center. A small team sits in front of walls of screens, scanning dashboards, reviewing alerts and logs, watching live feeds, and turning all of that activity into reports, often while waiting for something to demand attention.
Physical AI can change the role of that interface by moving intelligence closer to the equipment itself. A transformer station can recognize electrical patterns that may signal an emerging safety risk, and a sorting line can identify signs of an impending failure before operations are disrupted. The control room becomes less a place to watch every signal and more a place to coordinate decisions across the operation. Operators can focus on improving performance, planning maintenance, and acting before small issues become larger ones.
What if you could see the accident before it happened?
Not as a prediction or a crystal ball, but by recognizing the near misses and early patterns that usually disappear without a record. In the City of Bellevue, Washington, this idea is being applied to city streets by identifying repeated moments when vehicles, cyclists, or pedestrians come dangerously close, even when no collision occurs. Those moments reveal where risk is accumulating, giving the city a chance to redesign intersections, adjust traffic patterns, or improve street design before a serious accident occurs.
The same opportunity exists on a factory floor, at a construction site, or anywhere people and machines work together. A load swings past a worker. Two vehicles nearly cross paths. A machine behaves in a way that could have led to injury. Make those moments visible, and safety teams can improve the environment while there is still time to prevent harm.
And what if that intelligence could reach every asset in the field, not only the few that justify a custom monitoring system?
In oil and gas, utilities, infrastructure, and other distributed operations, fleets can include thousands of pumps, generators, compressors, and other assets across different makes, ages, environments, and levels of connectivity. Today, bringing intelligence to that equipment is expensive both to deploy and to maintain. Each asset may require its own instrumentation, definition of normal behavior, custom model, and deployment process, and every custom setup begins aging the moment it is finished (the challenge I explored in “The Machine That Won’t Sit Still”).
The ability to recognize behavior across many kinds of equipment changes that calculation. Intelligence can extend farther into the field, helping teams build a more complete understanding of the operation, improve performance and reliability, and extend the useful life of far more of their assets.
The opportunity is to build on what is already there. The PLCs, controllers, vision systems, and machinery developed over decades are the foundation for what comes next. Physical AI can add a new layer of understanding across them, helping people recognize what matters, act sooner, and bring intelligence into more of the physical world.
Imagine a workforce that spends less time watching for problems and more time improving performance. Imagine cities that can learn from near misses before they become accidents, and field operations where every asset contributes to a clearer understanding of what is happening.
The electric motor mattered not only because it replaced steam, but because it made a different kind of factory possible. Physical AI offers a similar kind of opportunity: not simply to operate within the systems we have more effectively, but to rethink what we build once intelligence no longer has to be centralized.
Draws on Paul A. David, “The Dynamo and the Computer” (American Economic Review, 1990).
Sisinio Baldis is head of solutions engineering at Archetype AI, where he works with teams deploying physical AI in factories, industrial sites, and other real-world environments. Physical AI from the Frontline is a field guide to what works, what doesn't, and why. New issues arrive every few weeks.