There is a car plant in South Carolina where, for most of last year, a robot turned up and did a job.
Not a demo. A real job. It loaded sheet metal into a welding line on ten-hour shifts, Monday to Friday, at an eighty-four second cycle time. Over ten months it handled more than ninety thousand parts, logged over twelve hundred hours of runtime, walked something like two hundred miles across the shop floor, and contributed to the production of more than thirty thousand BMW X3s with placement accuracy above ninety-nine percent. This June a newer model arrived at the same plant for a different job in logistics, sorting components into trolleys so they arrive at the line in the right order.
You probably didn't hear about it.
That isn't a failure of attention on your part. The attention went where attention always goes, which is to the demo videos. Robots doing backflips. Robots folding laundry badly. Robots being switched off by a nervous engineer. Those clips are designed to be watched, and they train you to file the whole subject under "impressive, not yet real." Meanwhile the unglamorous version, the one where a machine clocks in and builds cars for a year, gets a press release nobody reads.
So the story most people are carrying around is that humanoid robots are a decade away. The machines were already working while that story was being told.
What I got wrong about the pace
I've been publishing Global Macro Investor since January 2005, which is twenty-one years of writing down what I think is going to happen and then finding out. The job has always been the same one. Live a bit ahead of the tape, and be honest when the tape gets there before you do.
On this subject the tape has beaten me repeatedly.
I named robotics years ago as one of the curves that mattered, and I still underestimated how fast it would move. I wrote to my subscribers at the end of 2023 that even I had been shocked by the result, and I've had to write some version of that sentence most years since.
Robots are demographics
That's the line I keep using, and it's the load-bearing idea underneath everything below, so let me say plainly what it means.
An economy grows in exactly three ways. More workers, more output per worker, or more borrowing. Population plus productivity plus debt. I go through the full machinery of this in my Everything Code piece, and I'd rather point you there than rebuild it here, but the short version is that the first engine died decades ago. Birth rates across the developed world fell below replacement and stayed there. The workforce is shrinking. An older population produces less per head and draws more out.
You cannot fix that. The workers you would need were supposed to be born twenty years ago and they weren't. Every government on earth understands this, which is why every government on earth keeps borrowing. Debt has been doing the work of the two dead engines, and that borrowing is why your money loses purchasing power at roughly eight percent a year, which is a number I've written about at length elsewhere.
Here is what changes. You can't manufacture humans. You can manufacture labour.
A humanoid robot that can do physical work, and an AI agent that can do cognitive work, are not tools in the way a spreadsheet is a tool. In purely economic terms they are new productive units. They show up in the first term of that equation, the one everybody had written off as permanently dead. The formula stops being population plus productivity plus debt and starts looking more like humans plus robots plus synthetic labour plus debt, with energy yield and compute efficiency setting the ceiling on how fast the whole thing can scale.
That's the only exit from the debt trap that doesn't involve a default or a currency event, and it's the reason I find this subject hopeful rather than frightening.
The receipts, and the honest version of them
The manufacturing is already underway but it’s not distributed the way you'd guess.
One company in Hangzhou, Unitree, shipped more than fifty-five hundred humanoid robots last year. Over the same period Figure and Agility Robotics each shipped somewhere around a hundred and fifty.
Tesla has never disclosed a single production Optimus unit. The Model S and X lines at Fremont were shut down in May and are being converted, with production expected later this year. Musk has called it the hardest product Tesla has ever tried to scale.
The intelligence layer is Western. The best robotics foundation models are coming out of Physical Intelligence, Google DeepMind and Figure, and almost every humanoid on earth, including the Chinese ones, runs on Nvidia's chips and software.
The body is not. A humanoid is mechanically a bag of electric motors, twenty to forty of them, and high-performance motors need high-performance permanent magnets. China accounts for ninety-one percent of global rare earth refining and ninety-four percent of the world's sintered magnet production. In 2005 it was around half. That's not limitation of geology. Rare earths aren't rare. What's rare is the willingness to run separation chemistry that is technically brutal and environmentally filthy, which the West offshored on purpose because it was cheaper to let someone else do the dirty part.
source: iea.org
So we have an oddly clean separation. America is building the brain, and China is building the body.
Where machines are already deployed at scale, the productivity arrives exactly where the framework says it should. Amazon runs around a million robots. In 2016 an Amazon employee handled roughly 175 packages a year, and by 2025 that figure was closer to 3,870. That curve wasn't produced by cutting the warehouse floor. What changed is what one person plus a fleet of machines can do in a year, which is the actual definition of productivity growth.
That is the demographic hole starting to fill.
A new species, and we haven't noticed
We are not deploying equipment. We are creating a new life form and preparing to live alongside it, roughly the way Homo Sapiens lived alongside Neanderthals for five or ten thousand years. That sounds dramatic until you follow the economics, at which point it becomes the only framing that survives contact with the details.
Start with the money, because that's the part that arrives first and the part few people have thought through.
Robots and agents consume electricity and compute. Those costs have to be paid. An agent that books your travel, negotiates with three other agents and settles the bill is performing economic work and consuming resources to do it, which means it has to be remunerated. And the payments involved are instant and tiny, thousands of them an hour, between counterparties that are not human. Two-day bank settlement doesn't work for that. Card rails don't work for that. Venmo doesn't work for that.
So these things need to hold value, receive value, and transfer it without a person in the loop. Which means they need accounts. Which means, functionally, they get economic agency.
And here is the part that follows and that people flinch from. Economic rights precede social rights. They always have, in every expansion of the franchise in history. When something becomes an economically productive participant, the rest tends to follow, either granted or demanded. I'm not telling you that's desirable. I'm telling you it's the pattern, and I can't find a version of this where it doesn't eventually come up.
Then there's the strangest constraint, and it's Elon who saw it first and clearest. If you want to sell tens of millions of humanoid robots, people have to like them. A robot whose worldview doesn't reflect yours is a robot you find threatening and refuse to have in your house. Which is why he cares so much about what these systems are trained on, and it's a far more practical concern than the philosophical fights make it sound. You are not shipping a product. You are introducing something into people's homes that they have to be willing to live with.
Now stack it. We are putting rapidly improving intelligence inside a physical body that will eventually be stronger and more capable than we are, giving it the means to earn and spend, and designing it to be likeable enough that we accept it.
Whatever that is, it isn't a forklift.
The obvious objection
The obvious objection is that I'm anthropomorphising a machine. That a system predicting the next token is not a mind, a servo is not a muscle, and dressing statistics in a body doesn't make it a person.
That may well be right. I genuinely can't rule it out and I'm suspicious of anyone who says they can, in either direction.
But notice that the objection is metaphysical and the consequences are not. Whether these things are conscious changes almost nothing about what I've just described. They will still work. They will still need to be paid for the resources they consume. They will still transact at machine speed on rails humans can't service. They will still have to be built in a way we're prepared to accept. Every one of those pressures operates identically whether there's anybody home or not.
The philosophy is fascinating and I don't have the answer yet. All I know is this is coming fast.
What stays ours
Which leaves the question underneath all of it, the one that isn't really about GDP at all. If they do the work, what are we for?
David Mattin, who I write The Exponentialist with, has the best answer I've come across. He calls it the Creature Residual, deliberately echoing Solow's residual, the thing economists found in the numbers in the 1950s that couldn't be explained by labour or capital and turned out to be the thing that mattered most. His argument is that a post-human economy will have its own residual: the value only an embodied, mortal, conscious being can generate, which no machine can produce because producing it requires being one.
You can see where it concentrates. The premium people pay for a human therapist over a synthetic one. Live events, hospitality, care, mentoring, anything where being a person is inherent to what's delivered. Art most of all, where the entire point is that another consciousness made this and you're in contact with it across time. None of it gets cheaper as machine output gets cheaper. It gets scarcer, and therefore dearer.
David also thinks this eventually splits, between those who want to lean all the way into the technology and those who want to defend embodied human life. He calls the second group the creatures. What he's noticed, and I think this is the sharpest thing in the whole framework, is that the creatures aren't being sentimental. They're the custodians of the only asset the machine economy can't manufacture.
Where this leaves you
What I'll say is that the questions worth sitting with are not the ones getting asked. Not whether the robots are coming, because a machine in South Carolina already built thirty thousand cars last year. Not whether they take the work, because that was always the point and it's the only exit from the debt arithmetic I can find.
The questions are what we owe something that works and earns and might, for all we know, experience it. What we become when the thing that pays us stops being labour. And what remains scarce when nothing else is.
I don't have those answers. I'm not sure anyone gets to have them yet.
The machines will run the economy. Good for them. They'll do a wonderful job.
We mere creatures will hold the meaning.
David and I are working through all of it, in public, at theexponentialist.substack.com.
