Sometime late last week, in a livestreamed contest at Figure AI's headquarters, a visualisation-specialist intern named Aime Gerard out-sorted a humanoid robot called Bob the Bot by 192 packages over ten hours. Gerard finished with 12,924 parcels at 2.79 seconds each. The robot managed 12,732 at 2.83 seconds. Figure AI ran the challenge under California labour law, with paid breaks for Gerard, and according to CEO Brett Adcock the robot operated fully autonomously throughout. Gerard's left forearm was, in Adcock's words, "basically broken" by the end. "This is the last time a human will ever win," Adcock posted on X. It is a striking line to read in the same week that Hyundai disclosed plans for 25,000 humanoid robots, Barclays projected 24 million in China by 2035, and Adcock's other startup raised $700 million for AI hardware.
Hyundai's announcement is the most concrete of the three. The carmaker told JPMorgan investors it will deploy more than 25,000 Atlas humanoids inside its own manufacturing ecosystem, starting at the Georgia Metaplant in 2028 and expanding to Kia's Georgia plant the following year. It is targeting an annual production rate of around 30,000 Atlas units by 2028 and has committed to building more than 300,000 actuators per year at U.S. facilities. The robots are the same Atlas platform that Boston Dynamics, Hyundai's U.S. robotics arm, demonstrated this week lifting and carrying a 100-pound loaded fridge using whole-body control and millions of hours of GPU-based simulation training. Hyundai is no longer presenting itself strictly as a carmaker. It is positioning as a mobility and AI company that happens to assemble vehicles.
The Barclays report, picked up by eWeek, goes further on the demographics. China's working-age population fell from above 70 percent of the total a decade ago to around 61 percent, and Barclays expects its labour force to shrink by 37 million over the next decade. Under an optimistic scenario, humanoid robots could offset 60 percent of that decline by 2035 and equal nearly 4 percent of the labour force. China, the bank notes, already accounts for 85 percent of humanoid installations. The framing is no longer the science-fiction one of robots taking jobs. It is the actuarial one of robots filling jobs that there will not be enough humans to do.
Against that, the Figure AI livestream sits awkwardly. Ayanna Howard, dean of engineering at Ohio State, told reporters the robots' ability to work long shifts without failure was impressive, but said the technology is not ready for large-scale deployment. She cited specific problems: packages placed barcode-side up, boxes knocked off conveyor belts, accuracy gaps that would matter in a real logistics centre. One online commenter pointed out that Bob the Bot dropped roughly four times as many packages as Gerard and damaged more labels. The robot lost a single foot race against a human worker, on a task it was specifically designed for, by 1.5 percent. That is closer than it has ever been. It is also still losing.
Adcock's other bet is what makes the contest more than a marketing stunt. His new startup Hark, founded late last year with $100 million of his own money, this week raised $700 million in a Series A at a $6 billion valuation, with Nvidia, AMD Ventures and Salesforce Ventures in the round. Hark is building "personalised AI systems and hardware," with Apple iPhone Air designer Abidur Chowdhury aboard, and Adcock says Hark's AI agents are central to scaling Figure's humanoids. The argument, in effect, is that the gap closing in the package-sorting livestream is a software problem, and that scaling humanoid deployment is gated on the same model and data work that made chatbots competent at writing.
That is the open question. If Hyundai really has 25,000 Atlas units running on factory floors by 2029, and if Barclays' 24-million figure for China by 2035 is even roughly directionally right, the calculus of industrial labour changes in ways that few existing institutions are ready for. If the gap closes more slowly, the next decade will look more like advanced automation supplementing human workforces than replacing them. Aime Gerard, by 192 packages, briefly held back the line on one side of that question. The companies building Atlas, and the bank modelling 24 million Chinese humanoids, are committing serious capital to the bet that the next intern will not.