On a recent afternoon at London’s Brainstorm AI conference, Ty Brady, Chief Technologist of Robotics at Amazon, took the stage with news that could redefine warehouse automation, logistics, and the role of AI in the physical world. Brady unveiled “Vulcan,” Amazon’s first robotic system equipped with a sense of touch—ushering in what he described as “the gates of physical AI.” The move, highlighted in a session hosted by Fortune Magazine, signals Amazon’s mounting efforts to bring what the research community refers to as “embodied AI”—artificial intelligence that doesn’t just think, but acts in the physical world—to a massive scale.

Physical Intelligence on the Warehouse Floor

For years, Amazon has been a proving ground for the marriage of AI and robotics. Its fulfillment centers are already animated by the world’s largest fleet of mobile industrial robots, a fact Brady cited with proprietary pride. According to Brady, a staggering 75% of all Amazon deliveries pass through these robotic systems. Yet, the leap to robots endowed with a sense of touch represents a profound technical shift.

Unlike traditional robots that rely on pre-programmed routines or narrowly specialized automation, Vulcan integrates both vision and tactile sensing. “It will feel its way,” Brady explained, describing how the robot reaches into densely packed storage pods packed with hundreds of millions of different products. Utilizing force and torque sensors along with pressure-sensitive grippers—akin to a robotic “spatula”—Vulcan can both identify and extract objects in cluttered, unpredictable environments that had long stymied automation.

The technical breakthrough, Brady said, comes as much from advances in AI as from innovations in hardware design. While large language models train on vast corpora of online text to produce conversational intelligence, Amazon’s embodied AI systems amass enormous datasets from the trajectories, grasps, and manipulations required by warehouse work. “Data is the fuel for any AI system,” Brady affirmed, noting that every pick-and-place action feeds back into proprietary foundation models that refine and propagate robotic skills fleetwide.

Augmenting Humans, Not Replacing Them

The rise of increasingly dexterous warehouse robots has stoked perennial fears over automation and job loss. Nowhere is this debate more charged than at Amazon, the U.S.’s second-largest private sector employer. Yet, Brady was unequivocal: “We augment, extend human capability by having robotics be a tool set for people to use that allows them to be more operationally efficient… and provide a safer environment.”

He pointed to the company’s own data, which attributes a 34% reduction in repetitive injuries over the past five years to the deployment of robotics. Repetitive stress—often from reaching or bending for items stored on high or low shelves—remains a leading cause of musculoskeletal injuries in warehouses. By offloading the most ergonomically demanding tasks to robots like Vulcan, Amazon aims to keep more workers in what it calls the “power zone,” minimizing the risk of strain.

The allocation of work between robots and humans is dynamic—and deliberate. Around 75% of Amazon’s products, Brady attested, can be handled by Vulcan-style systems. “25%…is still a pretty big number, and there are some objects where people are just going to be better at [handling them],” he said. Those edge cases are routed to human workers, who focus on exceptions, orchestration, and complex tasks beyond the reach of current-generation robots.

Impact on Amazon’s Workforce

If the narrative at Amazon is “employees first,” questions remain about the long-term trajectory of warehouse labor. Will automation ultimately shrink headcount? Not necessarily, argued Brady. In Amazon’s latest next-generation fulfillment center in Shreveport, Louisiana, an investment in ten times more robotics than previous facilities did not reduce jobs; it created 2,500 new positions. Approximately 30% are “more technically focused,” buoyed in part by Amazon’s stated $1.2 billion commitment to upskilling employees for tech roles—from mechatronics apprenticeships to technology education via its “Career Choice” program.

Brady sketched a “collaborative robotics” flywheel, where robotic productivity lifts throughput, enables greater selection and lower consumer prices, and unlocks yet more opportunities for workers. He flatly rejected the notion that robots cannibalize jobs: “You could be [at] a thousand or more jobs, in fact,” he said of high-automation sites, noting that higher productivity yields more units, more exceptions, and more supervisory and technical roles for humans.

Another critical component of embodied AI is its capacity to learn at scale. Brady detailed how every object manipulated not only trains the handling robot, but also improves the foundational model for its entire robot fleet.