This week Brett Adcock, the chief executive of Figure AI, posted a chart that crossed a line many people in robotics have been waiting years to see. For the first time, his company's robots outnumber its employees: roughly 740 humanoid units against 650 to 660 staff. The number is partly a vanity metric, because the robots are mostly being built for customers rather than put to work inside Figure itself. But as a marker of where the industry has arrived, it is hard to beat. The thing being manufactured now multiplies faster than the people manufacturing it.
That milestone landed as Automate 2026 opened in Chicago, the largest edition in the trade show's 50-year history, with more than 50,000 attendees and, for the first time, a dedicated Humanoid Robot Pavilion sponsored by NVIDIA. The framing has shifted in a way that organisers were happy to say out loud. The question is no longer whether humanoid robots are ready. It is how quickly industry can absorb them now that they are. Three deployments anchor that confidence: Agility Robotics' Digit working commercial shifts at a Toyota plant in Woodstock, Ontario under a leasing contract; Boston Dynamics shipping its electric Atlas, with the entire 2026 run committed to Hyundai and Google DeepMind; and Figure's own production line hitting one robot per hour.
Underneath all of it sits a single phrase repeated across the show floor: physical AI. The idea is that robots learn tasks by watching humans perform them rather than by being programmed step by step. Standard Bots' Evan Beard will argue in a keynote that conventional automation only ever reached about 1 percent of factory tasks, because writing explicit code for every variation was too expensive. Demonstration-based learning is the bet that the other 99 percent is finally reachable. The business model arriving with it is Robots-as-a-Service: rather than buying a machine outright, manufacturers rent it monthly, which turns a daunting capital purchase into an operating cost and, for the first time, puts humanoids within reach of smaller firms.
The commercial wave is already visible in less glamorous corners. Robot.com, formerly the delivery startup Kiwibot, launched a wheeled humanoid called R-noid this week for packing, picking and folding, and says it can go from a customer's first site visit to autonomous operation in eight to twelve weeks. NVIDIA used the show to introduce Halos, a safety system it bills as drawing on 18,600 engineering-years of self-driving-car work. And in Hong Kong, a pop-up convenience store staffed by a single humanoid opened on the waterfront, a reminder that the same technology arrives as both industrial tool and street-level curiosity.
What the celebration tends to skip is the gap that every forum session keeps circling back to. A robot that succeeds 70 percent of the time is a great demo and a useless colleague; production lines need reliability above 99 percent, and the sim-to-real gap, the way a policy trained in simulation degrades on a real floor with real friction and clutter, is not solved. Nor is the safety question of letting robots work alongside people without fences, where the certification standards are still being written. The robots now outnumber the humans at one company. Whether they can be trusted to do the work, every hour, beside the humans who remain, is the part that 2026 has started rather than settled.