Boston Dynamics hands Atlas's move into factories to former Amazon AI chief Rohit Prasad

RuntimeWire describes the job as turning technical performance into reliable, repeatable work that customers will pay to scale ^[13]. The outlet links the timing of the hire to Atlas's move toward production ^[11]. That move changes what counts as success. A controlled demonstration has to work once. A deployment, by RuntimeWire's account, has to run safely and reliably across shifts, fit into the plant's workflow and compete with the automation already installed ^[14].

The most recent Atlas step the outlet covered was the robot carrying a mini fridge, a test of whole-body manipulation with an awkward object ^[18]. It was a sensible choice of task, because it put the emphasis on manipulation over acrobatics ^[18]. But the reporting does not include a reliability figure, a unit price or a customer beyond Hyundai and Google DeepMind ^[10].

Prasad has said the talking computer on Star Trek helped steer him toward speech processing and conversational AI ^[19]. His actual record is scale in software. Boston Dynamics says Alexa reached hundreds of millions of households ^[8], and Prasad later led Nova, Amazon's foundation-model family ^[2]. Those are company-reported measures of distribution. For them to predict how Atlas will do, the hard part of Atlas would have to be its software. RuntimeWire's counterpoint is that robotics adds hardware, safety, service and integration demands that a voice assistant does not face ^[15].

I think the board is betting that the software is the hard part. Boston Dynamics calls its effort Physical AI, meaning systems that perceive and act in real environments ^[9], and a foundation-model executive fits that description of the problem. The company's roots are in hardware. Founder Marc Raibert started it after research into dynamic, legged robots ^[12], and its work has long been judged by what its machines can do in motion ^[17].

Hyundai has said it plans to deploy Atlas at its Georgia plant beginning in 2028 and to build capacity to produce as many as 30,000 humanoid robots a year, Bloomberg reported ^[4]. Counted from the 2026 deployments, that is about two years ^[20]. The 30,000 is a capacity target. It does not describe current production or sales ^[4]. RuntimeWire calls Hyundai a potential first customer and test site, and names demand beyond Hyundai and economical production at scale as the tests still ahead ^[16].