Boston Dynamics hands its keys to Amazon's foundation-model chief

For three decades, Boston Dynamics was run by people who understood legs. Its new chief executive spent the last few years understanding language models. On October 7, the Hyundai-owned robot maker named Rohit Prasad, formerly senior vice president and head scientist for Alexa and artificial general intelligence at Amazon, as its CEO, Reuters reported. He is only the third person to hold the job since the company spun out of the MIT Leg Lab in 1992.

Prasad replaces Robert Playter, who stepped down in February after six years as CEO and more than 30 years at the company. CFO Amanda McMaster ran things on an interim basis during the search, according to Robotics 24/7 and Manufacturing Dive. Prasad is expected to join the board once approvals clear.

The hire is not subtle. Boston Dynamics is famous for hardware: Spot, Stretch, and the Atlas humanoid, whose videos have defined what a capable robot looks like for a generation of engineers. Choosing a foundation-model executive to lead it is a statement about where the company thinks the remaining problem lives.

Who Prasad is

Prasad spent 12 years at Amazon. He was a central figure in building Alexa into a household product, then took over Amazon's AGI group in 2023, where he led development of Amazon Nova, the company's family of foundation models aimed at enterprise customers. Before Amazon, he spent nearly 14 years at Raytheon BBN Technologies leading machine learning research.

His Amazon exit was announced last December. Fortune reported on December 17, 2025, that CEO Andy Jassy was folding the AGI unit into a larger organization under AWS veteran Peter DeSantis, combining it with custom silicon and quantum computing. Jassy called Prasad "missionary, passionate, and selfless" on the way out. Neither Amazon nor Prasad said at the time where he was headed.

In the Boston Dynamics announcement, Prasad said the company "is uniquely positioned to advance Physical AI through its world-class robotics expertise." Jaehoon Chang, the Hyundai Motor Group vice chair who chairs the Boston Dynamics board, credited Prasad's "proven ability to turn AI into products at global scale." That phrase is the job description. Not invent new robots. Ship intelligence into the ones that exist.

The company already admitted the gap

The tell came at CES in January. Boston Dynamics walked the production Atlas onstage, called it the best robot it had ever built, and in the same breath announced a partnership with Google DeepMind to bring its Gemini Robotics models to the machine. A company that had spent decades writing its own control software went outside for the reasoning layer.

That decision explains October's hire better than any press release. The body problem is largely solved, or at least on a manufacturing roadmap. The electric Atlas shown at CES reaches about 7.5 feet and lifts up to 110 pounds, according to SiliconANGLE. Atlas general manager Zachary Jackowski said every component was designed for compatibility with automotive supply chains.

The hardware keeps moving. On October 1, six days before the CEO news, Boston Dynamics unveiled a new four-finger hand for Atlas with 13 degrees of freedom, up from 7 on the previous hand, according to The Robot Report. The team dropped the pinky after deciding the extra dexterity was not worth three more degrees of freedom and the added size, power, and actuators. Dense tactile sensors cover the fingertips and palm, and demos showed grips on drills, grinders, nail guns, and welding torches.

Even that hand is a software story in disguise. The company said it designed the rigid, backdrivable mechanism specifically so it could be simulated with high fidelity, and argued that reinforcement learning in simulation is essential to dexterous manipulation. Outlets disagreed on how much of the demo footage was teleoperated versus autonomous, which is itself the point: the gap between what the hand can physically do and what Atlas can decide to do on its own is the gap Prasad was hired to close.

The clock Hyundai has set

Hyundai bought out SoftBank's remaining stake in July and now owns the company outright. Its plans are aggressive. Reuters reported that the group targets a robot factory with 30,000 units of annual capacity by 2028 and intends to begin deploying robots across its manufacturing sites that year. Manufacturing Dive put the planned Atlas fleet at 25,000 units across Kia and other plants. The figures differ by source and both are targets, not commitments in hardware.

Per the CES announcements, the first Atlas units are slated for Hyundai's EV plant near Savannah, Georgia, starting in 2028 on parts sequencing, with vehicle assembly tasks to follow around 2030. Boston Dynamics opened a robotics center at that Georgia site in September.

That pressure has costs. Manufacturing Dive, citing Semafor's May reporting, said the chief operating officer, the chief strategy officer, and a number of senior engineers and researchers left after Playter's departure, with former employees describing pressure to integrate tens of thousands of humanoids into Hyundai's plants within a few years. Prasad inherits both a deadline and a thinned bench.

What a model executive actually changes

The practical question is what Prasad's background buys. Building Nova meant managing data pipelines, training compute, evaluation, and the long, unglamorous path from benchmark demo to a product enterprises trust. Those are exactly the muscles a humanoid program needs once the actuators work. A factory robot is judged on uptime and error rates across thousands of shifts, not on a viral clip.

There is an open question about how Prasad squares the DeepMind partnership with any ambition to own the model layer in-house. Boston Dynamics has not said whether Gemini Robotics remains the long-term brain for Atlas or a bridge while the company builds its own. A leader who spent years building foundation models at Amazon is an unusual choice if the plan is purely to license someone else's.

What to watch

The appointment reframes the humanoid race. For years, the assumption was that whoever built the best body would win, and Boston Dynamics had the best body. Its owner has now concluded, with a CEO hire rather than a blog post, that the body is table stakes. Watch three things: whether Boston Dynamics announces its own robot foundation model or deepens its reliance on DeepMind, whether the departures stop, and whether Atlas units in Georgia in 2028 are doing autonomous work or waiting on a remote operator. The answer to the last one will say whether this bet was right.