After years of steady progress, robotics is entering a more consequential phase, and the shift is coming exactly when it’s needed. By some estimates, nearly two million manufacturing jobs could go unfilled by the early 2030s; warehouse turnover remains stubbornly high. The surge in robotics funding over the past three years reflects the growing belief that robotics can help fill these gaps and take workplace productivity to new heights.
This transition from promise to practice framed the panel discussion on “Transforming industries with physical AI,” moderated by McKinsey partner Ani Kelkar at CES 2026 in Las Vegas. The discussion convened leaders from Google DeepMind, Boston Dynamics, General Motors, and Qualcomm, and each panelist grappled with the same questions: What is accelerating robotics adoption? How quickly will the industry progress over the next few years? And what will it take to scale robotic deployment?
According to Carolina Parada, senior director and head of robotics at Google DeepMind, recent growth reflects advances in robots’ perception and interpretation capabilities. “Five years ago, robots didn’t really understand their environment,” she said, “You had to essentially code every single step.” Today’s foundation models combine vision, language, and reasoning skills, allowing robots to follow natural-language instructions and adapt their actions in real time. Some of the advances—robots doing laundry, for example, or having the dexterity to tie shoelaces—have surprised even industry experts. “These things were once considered impossible,” Parada said.
Several panelists emphasized that robotics should be understood not as a finished machine but as a long-lived platform—and adoption timelines may vary by application. Nakul Duggal, senior vice president and general manager of automotive, industrial, and embedded IoT at Qualcomm Technologies, noted that, over the next five years, companies will continue to deploy long-standing robotics applications while also pursuing new use cases. “The technology adoption cycles can be quite different because the problem you’re trying to solve is not the same,” he explained.
Mikell Taylor, director of robotics strategy at General Motors, reinforced that progress varies by technology. Highly regulated, high-reliability environments such as automotive manufacturing demand predictability above all else. “A lot of technologies that have been around for ten years are ready for those environments now,” she said. “But some of the more modern AI embodied in robots still has a long way to go in development before they’re ready for those environments.”
Parada cautioned against thinking of readiness as a binary state. “It’s not that AI will suddenly be ready for full commercial deployment all at once,” she said. Instead, AI will be layered into existing automation, improving perception and diagnostics while gradually unlocking new applications. “We’re not just making existing robots ten percent more efficient,” she said. “We’re enabling robots in places we never even considered before because the variability was too high.” That shift, however, comes with new challenges. “Using AI as a physical system is very different from using it as a chatbot,” Parada noted, pointing to safety, standards, and accountability.
Robert Playter, CEO of Boston Dynamics, believes that the industry’s center of gravity has already shifted significantly. “For the 30 years I’ve been doing robotics, it’s been a niche, and it’s been dominated by YouTube videos and prototypes,” he said. “The interesting evolution is that scaled adoption is beginning to happen.” Playter also noted that more companies are creating robotics components such as actuators, batteries, and power systems—the “industrial might” required for sustained growth.
As companies investigate new use cases, they should focus on challenges related to an aging workforce and declining interest in physically demanding jobs. “The next decade is going to be about identifying jobs that simply should not be performed by humans,” Duggal said. Many future roles, he suggested, will involve robots operating autonomously and bringing humans into the loop when judgment is required. Opportunities span healthcare, hospitality, services, and defense.
In automotive manufacturing, Taylor noted that body and paint shops are heavily automated but final assembly still remains largely manual—but that could soon change. “Tasks like installing wire harnesses or fitting interiors require robots to recognize subtle variations across models and adapt in real time,” she said. “These are precisely the kinds of challenges advanced AI is beginning to address.”
Parada argued that researchers could help robots master new applications if they rethink the training process. Rather than training systems on narrow, domain-specific routines, her team focuses on fundamental motions and feedback that are applicable in multiple settings. “If you just specialize, you end up with a brittle model,” she said. Exposure to diverse tasks builds robustness—and placing robots in variable environments, such as automotive assembly, may actually strengthen learning. The new models alone will not automatically translate into industrial readiness, Parada noted, since progress depends on advances in hardware, sensing, and data—not a single breakthrough.
The panelists repeatedly emphasized that trust will be a decisive factor in scaling robot deployment. When robots are introduced into a workplace, Taylor noted, people are often reluctant to delegate tasks to them and managers do not trust the technology. “Those first few years are a critical period to get it right—and it’s not engineering; it’s thinking about the human,” she said.
Playter agreed. “People’s attitudes don’t scale as fast as the tech,” he said. In his experience, companies often need two to three years to get used to having robots in their facilities before they are ready to expand—and that delay is rarely factored into projections that assume exponential growth. Boston Dynamics, he said, is beginning testing with Hyundai and expects more advanced pilots in 2027, with factory deployments around 2028. That timeline reflects not just technical readiness but organizational readiness. “Going a little slower leads to a better outcome,” Playter concluded.
Looking at the bigger picture, Duggal emphasized the importance of a global ecosystem that encompasses all the important elements, including hardware, software, and data. “We’ve been working with companies all over, from Japan to Germany, and I think there is a new supply chain for robotics emerging,“ he said. The robotics ecosystem is evolving so quickly that it may look very different in five years.
Building the right talent is also imperative for future success, the panelists agreed. “The mechanical and electrical engineering skills haven’t changed, but the AI skills are new and the method for programming the robots has largely been transformed in the past few years, so there’s a huge talent war going on right now,” Playter noted. He was optimistic, however, that supply and demand might find a better balance in a few years. Taylor said that upskilling employees at every level will be critical to filling future talent gaps in development, integration, and operations. “If you have to compete for all that talent externally, that’s going to be a war no matter what,” she said.
Parada emphasized that companies must look beyond their engineering needs when considering essential skills and technologies. Every employee does not have to be a robotics expert; instead, companies should ensure that the robots and models are user friendly. “You shouldn’t need to be an AI specialist to work with these systems,” she said.
***
The panel ended on a note of realism. Flashy demos are easy; scaling adoption and finding valuable use cases is hard. Although challenges remain, the industry is headed in the right direction, and robots are not doomed to languish in small pilots. Over the next five years, the defining question will not be whether robots can work but how organizations can gain the most value from them.