For as long as anyone reading this can remember, we have been promised that the general-purpose humanoid robot is arriving next year, and next year has reliably declined to show up. In July 2026, something actually shifted: the robots stopped performing choreographed stage demos for the investor class (or increasingly kicking people, as dozens of viral videos have shown lately) and started clocking in for real factory shifts.
The world’s richest man and cartoonish “Sigma Male poster boy” Elon Muck… sorry, Musk, is ramping production of Tesla’s third-generation Optimus (so cool, Elon!) at its Fremont plant. It now packs an in-house AI5 chip (around five times the memory bandwidth of the previous one) and, because it is a Musk production, xAI’s Grok bolted in to handle the voice, a pairing that is either the future of human-robot interaction or three class-action lawsuits stacked on top of each other in a trench coat, depending entirely on your taste or lack thereof.
And Tesla is not alone in slow-cooking the future employment prospects of the ever-growing underclass.
Figure, the buzziest of the pure-play robotics startups, says its Figure 03 units are now rolling off the line at roughly one robot per hour, and it already has them doing sequencing work at BMW’s Spartanburg plant in South Carolina.
The catch, and with robotics there is always a catch, is that you still cannot buy one of these things. No consumer pre-orders, no waitlist, nada, and analyst consensus has slid from “end of 2027” to a far more realistic 2028 or 2029 (the humanoid timeline, much like the self-driving one that came before it, has a genuinely remarkable talent for staying perpetually two years away).
Figure is winning the part that counts
Strip away the Musk theatrics and Figure is quietly running ahead where it matters: actual paying customers. Its BMW deal is not a demo, it is a commercial contract covering an initial fleet of forty robots doing logistics in Hall 52, priced at roughly $25 per robot-operating-hour on a robot-as-a-service basis, after a ten-month pilot with the previous model produced tens of thousands of parts. Tesla, by contrast, has vastly more cash, a chip roadmap, and Fremont’s manufacturing muscle, but as of this spring it had exactly zero announced external customers. One company is shipping robots that do work today; the other is promising to ship better robots to more people later. Which of those bets pays off is the entire game.
The AI tool stack actually worth paying for
One email a week. The models, tools and moves that matter, stripped of hype and filtered so you don’t have to drink from the firehose. Free, and you can bail anytime.
The part the promo videos skip
Here is the uncomfortable subtext underneath all the “physical AI” excitement. The immediate, provable use case for humanoid robots is doing repetitive industrial labour that human beings currently do for a wage. Figure 03 is in that BMW hall precisely because it can pick and sequence parts that a person used to pick and sequence. That is genuinely impressive engineering, and it is also a slow-motion answer to a question nobody in the promotional videos wants to ask out loud: what happens to the humans currently holding those jobs when the robot works a full shift for twenty-five dollars an hour and never asks for a break.
For now, the honest state of play is this. The robots are real, they are in real factories, and they are doing real, boring, valuable work. They are also years away from your living room, wildly oversold by at least one CEO with a stock price to prop up, and quietly reshaping the bottom of the labour market while everyone argues about whether they will one day fold laundry. Manage your expectations, and maybe keep an eye on who benefits.