For a long time, the public image of advanced robotics was the viral video: a humanoid machine doing a backflip, dancing, or navigating an obstacle course. Impressive, shareable, and almost entirely beside the point. Those demos told us robots could move. They told us nothing about whether robots could work.
What has changed recently, and what I think marks a genuine inflection, is that humanoid and mobile robots are moving from the demo reel onto actual factory floors. Not to do anything spectacular. To do the dull, repetitive, physically taxing tasks that are hard to staff and harder to keep people in. The story has shifted from "look what it can do" to "look what it is doing," and that is a much bigger deal than another acrobatic clip.
The reason this is happening now is a convergence rather than a single breakthrough. The mechanical hardware has matured to the point of real dexterity and endurance. But the more important change is the brain. Advances in AI mean a robot can now learn a new task quickly, adapt to a messy and changing environment, and handle the ambiguity of the physical world rather than needing every motion pre-programmed. The intelligence that made AI useful in software is now being poured into machines that can act in the physical world.
I think we underrate how significant the physical embodiment of AI is. So far, the AI revolution has mostly lived on screens — generating text, images, analysis. That is powerful, but it leaves the entire physical economy largely untouched. The moment capable AI gets a body that can manipulate objects, move through space, and learn on the job, the addressable territory expands from information work to the vast world of physical labour. That is a step change in scope.
There are obvious limits, and I want to be honest about them. The physical world is unforgiving. The gap between a controlled factory environment and a chaotic, unstructured one is enormous, and many tasks that look simple for a human remain stubbornly hard for a machine. I do not expect humanoid robots to be everywhere quickly. The early deployments will be narrow, in environments that can be engineered to suit the machine.
But the trajectory matters more than the current state. We have crossed from "can it move impressively" to "can it earn its keep," and once a technology starts earning its keep, the incentives to improve it intensify dramatically. Each real deployment generates data and revenue that fund the next generation. That feedback loop is exactly what turned earlier technologies from curiosities into infrastructure.
There is a labour dimension here that deserves more honesty than it usually gets. The first wave targets work that is already hard to fill — jobs people increasingly do not want. That softens the immediate disruption. But the capability does not stop there, and the same machine that relieves a labour shortage today is the machine that competes for a job tomorrow. We should be thinking about that arc now, while it is still early, rather than after it has arrived.
I think the honest stance is to hold two truths at once. These machines will relieve genuine burdens — dangerous, exhausting, unfillable work that almost no one will mourn. And the same capability that relieves those burdens will, on a longer horizon, contest work that people very much want to keep. Pretending only the first is true is exactly how societies sleepwalk into the second, and then act shocked when it arrives.
The era of robots as entertainment is ending. The era of robots as colleagues is starting. The question worth sitting with is not whether they can do the work, but how quickly the set of work they can do expands once the loop starts turning — and whether our institutions are anywhere near ready for the answer.
The above reflects my personal views only and is intended for informational and discussion purposes. It does not represent the position of any employer or organisation.