For thirty years Boston Dynamics has been the company that makes robots do impossible things with their bodies. Its machines backflip, parkour, and sprint faster than Usain Bolt. So the most revealing thing about its new chief executive is that he has almost no background in any of that. On 7 October, Rohit Prasad, the former Amazon scientist who built Alexa, took over the robot company. The choice is the story: the hard problem at Boston Dynamics is no longer the body. It is the brain.
Prasad spent twelve years at Amazon as head scientist for Alexa and, later, its Nova foundation models, turning research into products used in hundreds of millions of homes. He is an AI person, not a hardware engineer, and the press release leans hard on a phrase the whole industry has adopted: "Physical AI," meaning systems that can perceive, reason, and act in the real world. He succeeds Robert Playter, only the second CEO in the firm's history, who stepped down in February amid what one report called a C-suite exodus.
The timing is not subtle. Parent company Hyundai wants a factory capable of building 30,000 humanoids a year by 2028, with 25,000 of them going to work in its own car plants and just 5,000 sold. Atlas robots are already training at a new center in Georgia, learning to sequence automotive parts. The ambition is enormous, and the current reality is small: Boston Dynamics was reportedly making only four Atlas units a month earlier this year, at an estimated 130,000 to 140,000 dollars each, a cost it hopes to crush to 30,000 only after it has built 50,000 of them.
Why reach for an AI executive to close that gap? Because the bottleneck has moved. As a sharp column from DigiTimes put it this month, robots have learned to move, but they have not learned to think. A dazzling demo of a dual-arm machine catching randomly dropped sticks is not the same as a robot that can handle an unfamiliar object on a chaotic production line. The body is largely a solved engineering problem. The intelligence to make it reliably useful, shift after shift, is not.
The sober evidence backs that up. The genuinely impressive deployments are narrow. Figure's robot worked ten months at a BMW plant, moving more than 90,000 parts, doing one repetitive task. Agility's Digit has shifted over 100,000 totes in a warehouse. These are real milestones, but they are a long way from a general worker, and the field's own experts keep stressing the gap between a pilot and a production deployment. Dexterity and the ability to generalize to new conditions remain the stubborn frontier.
Then there is the supply chain nobody in the launch videos mentions. China installed 59 percent of the world's industrial robots last year and controls roughly 90 percent of the refined rare earths and permanent magnets that make robot motors spin, a structural grip on the hardware that no amount of software brilliance erases. So Prasad arrives with a clear mandate and a hard ceiling. His bet, and Hyundai's, is that the missing ingredient is a mind. The question is whether the company that taught robots to run can now teach one to understand the factory it is running through.