The factory deployments don't tell you what the real bet is
Two car companies can put the same robot on the same factory floor and be betting on completely different futures.
BMW is paying Figure to put a humanoid on its factory floor. Mercedes is testing Apptronik's Apollo in its plants. Toyota is renting Agility Robotics' Digit on a subscription. Those companies aren't betting that humanoids become the next big product. They're betting a humanoid might be a better factory tool than the equipment they already have. If they're wrong, they write off a procurement decision and move on. That is an efficiency bet. The robot is a means to make the business they already have run a little better.
Apptronik’s Apollo at Mercedes-Benz factory. (Image Credit: Apptronik)
Then there's the other camp. XPeng didn't just put a robot in a factory. It built its own artificial intelligence chip that runs the robot's decision-making, repositioned itself as an embodied intelligence company and says it's building a dedicated humanoid factory with mass production targeted by the end of 2026. Chery is already selling its Moyin humanoid. Xiaomi is using its own humanoid inside its factories while developing the platform further. Tesla says it's reshaping part of Fremont around Optimus production with ambitions measured in millions of units. Li Auto's chief executive called this the "final opportunity" to become a leading global artificial intelligence company. That isn't an efficiency strategy. It's a product strategy. Cars fund the business today. Humanoids are supposed to become the business tomorrow.
Most coverage merges these stories because the robots look identical. But buying someone else's robot and trying to become the company everyone else buys robots from are fundamentally different wagers, with different risks if they fail. The sceptics arguing that purpose-built machines will outperform humanoids on factory floors may well be right, and that matters if you're BMW. It matters much less if you're XPeng, because their destination is the showroom. The factory is just the backdrop. The real question isn't which automaker adopts humanoids first. It's which one stops thinking of itself as an automaker.
Enjoying the read so far?
What's hiding behind the "world-first" headline
I kept seeing headlines about a "world first" in robotic surgery, as though the only question worth asking was whether this machine beat the last one.
That feels like the wrong comparison.
Surgical robots have been in operating theatres for about 25 years. They already perform delicate procedures every day. Against those systems, the new result is almost modest. In a Nature paper published this month, the UCSD team reports that the humanoid needed recalibration during the procedure, and it was slower than existing surgical robots. They also describe it as teleoperated, meaning a surgeon controlled every movement throughout.
The study was preclinical, meaning the operation was performed on pigs rather than people. None of that makes it look like the next operating theatre standard. It makes something else harder to ignore. A general-purpose humanoid arm can now carry out a laparoscopic gallbladder removal, where surgeons operate through small cuts using long instruments, while the person making the decisions stands somewhere else entirely. That shifts the question from performance to geography.
Two teleoperated Unitree robots work as a team to perform a laparoscopic gallbladder removal during a pre-clinical trial. Image Credit: University of California San Diego
Most discussions treated this as another entry in the endless contest over which robot is more precise. I couldn't stop thinking about who never gets compared because they never get surgery at all. Much of the world has no access to advanced surgical robots. Large numbers of people also have no access to specialist surgeons. Those two shortages usually arrive together. A machine that merely reaches parity with existing systems looks underwhelming if your benchmark is a flagship hospital. It looks different if your benchmark is a patient whose nearest specialist is hundreds of miles away.
The paper never claims to solve that problem. Coverage mentioned it, but didn't lead with it. The result is not that humanoids outperform dedicated surgical robots. The result is that a body designed to do many jobs can now become a pair of surgical hands somewhere else, provided a surgeon can reach it through a network connection. If the hardest part of surgery stops being where the surgeon is standing, what becomes the new bottleneck?
Editor’s Take
Both pieces this week are about what's underneath a headline. The automaker story looks like adoption; it's actually two different wagers on what humanoids are for. The surgery story looks like a benchmark; it's actually a finding about who surgery reaches.
It is the obvious which is so difficult to see most of the time.