The automated workforce: Physical AI's labor impact

Digital vs. physical economy

The digitalization of business has been a key economic growth engine for years, with AI providing the latest push. But hands-on labor remains essential amid the rise of physical AI.

"The thing that people miss is the physical economy is enormous," said Ani Kelkar, a partner at McKinsey & Co. "Despite all the attention we understandably give to the digital economy, 44% of paid work hours globally, across sectors, are tied to physical work."

Speaking during the McKinsey Live webinar, "Physical AI: The hidden value pools leaders are missing," Kelkar said that 13% of work hours can be automated today. The percentage of automatable working hours, he added, amounts to about 20 million full-time workers' worth of annual tasks that physical AI can address. "This doesn't mean that those jobs are going to go away," he said. "In fact, those jobs will likely get reinvented."

Labor augmentation vs. replacement

Rather than wholesale replacement, physical AI seems poised to operate alongside workers.

Joshua Morley, group chief AI officer at digital engineering consultancy Akkodis, said he expects augmentation to be the primary story, particularly for work that isn't completely linear -- where exceptions arise that require judgment or human interaction. "In these kinds of environments," he explained, "we're expecting physical AI to take on the repetitive, physically demanding or monotonous parts of the process, where the people can take responsibility for the outcome."

Physical AI will take on tasks that are repetitive, dangerous, highly structured or physically demanding, while humans remain responsible for judgment, supervision and exceptions.

Hantz Févry, founder and CEO of spatial intelligence platform provider Geolava, also sees physical AI playing a supplementary role. His company offers a world model for the built environment that learns how real-world assets change over time, enabling organizations to predict consequential physical changes.

A key distinction is that "world models dramatically expand the range of tasks machines can perform," Févry said. A world model gives machines something closer to an internal simulation of what might happen next, he explained, and that lets physical AI operate in environments that previously required human adaptability.

"Initially, I expect augmentation to dominate," Févry said. "Physical AI will take on tasks that are repetitive, dangerous, highly structured or physically demanding, while humans remain responsible for judgment, supervision and exceptions."

Warehouses, factories, construction sites, transportation and infrastructure inspections are obvious examples, Févry noted. "Over time, however, some categories of labor will absolutely be automated," he said. "I do not think we should pretend otherwise."

Direct labor replacement is likely to emerge in narrowly defined tasks that might be hazardous, highly repetitive, ergonomically challenging or simply undesirable, Morley surmised. Such tasks might involve night shifts or hazardous work such as underwater welding.

Physical AI drives new types of work

In some cases, physical AI will open opportunities for human workers to take on new workplace roles. Morley cited the use of autonomous haulage vehicles in Western Australian mines, where these vehicles created new maintenance, systems engineering and integration roles.

The result is safer jobs and increased productivity. In 2024, Komatsu delivered its 300^th autonomous haulage system to mining company Rio Tinto's Pilbara operations in Western Australia. At the time, a Komatsu official said the vehicles addressed mine safety and labor shortages, while "enabling continuous operation."

An Akkodis June 2026 research report, "What CTOs Think 2026: Scaling the agentic enterprise with confidence," suggested that AI is changing work rather than driving widespread job cuts. Half of the 500 CTOs surveyed for the report said AI changes the skills needed for certain roles, while nearly half said AI changes employees' day-to-day activities. A smaller slice, 21% of the CTOs, said AI reduced headcount. "So, it's mostly a reshaping of what the work is, rather than a direct replacement," Morley said.

Févry said he expects to see "entirely new forms of work emerge" from physical AI adoption. "Humans will increasingly manage fleets of intelligent physical systems, define objectives, supervise edge cases and orchestrate machines, rather than manually perform every task," he said.

With that in mind, the long-term shift is bigger than replacing individual jobs, Févry said. Humans will move increasingly from operating machines to managing intelligence embodied in machines, he noted.

Physical AI's evolution

The continued evolution of physical AI will change the technology's role in the workplace and where human workers encounter it.

The physical AI use cases that arrive relatively quickly will involve moving items around in a commercial context, for instance, forklifts, industrial inspection equipment and robot arms, said Mark Patel, senior partner at McKinsey, during the physical AI McKinsey Live webinar. "In the near term," he explained, "this is likely to be a lot about mobility and a lot about specialized equipment, because that's where the technology maturity and the commercial use case intersect."

In the near term, this is likely to be a lot about mobility and a lot about specialized equipment … where the technology maturity and the commercial use case intersect.

Physical AI will likely follow the broader pattern emerging around enterprise AI, said Anant Adya, executive vice president at IT services provider Infosys. The company's 2026 study "The AI ROI Gap: Turning Ambition into Enterprise Value," found that AI's strongest impact is on the operational side, especially in speed-to-market and cost savings, he said. But revenue gains are taking longer to emerge, he added.

"With physical AI, that suggests the first compelling business cases will often be tied to measurable improvements in how work gets done," Adya explained. "Quality inspection, predictive maintenance, asset monitoring and worker safety are good examples." These business cases connect AI to outcomes such as fewer defects, less downtime and lower operational risk, he said.

Those initial deployments will open the way for another class of applications. "As AI converges more deeply with robotics and other physical technologies," Adya said, "these early gains can create the foundation for enterprises to redesign processes and, ultimately, pursue new products, services and business models."

The more sophisticated applications will take longer to become mainstream. Use cases requiring greater precision in dexterity and more thoughtful perception of the environment might achieve higher accuracy, Patel noted, but they'll also involve a much more acute application of the models and technology. "Those are ones that we should expect to take longer to learn and to improve before we see them widespread," he said.

Robot dexterity and demos

The need for greater dexterity is an important issue in physical AI -- and an obstacle that currently limits what humanoid robots can achieve. A human hand, for example, has 27 degrees of freedom of movement. Robotic hands are approaching that mark, but they also need to embed finely tuned tactile capabilities.

A May 2026 paper, "Towards Robotic Dexterous Hand Intelligence: A Survey," cited "a fundamental trade-off between high degrees of freedom and system complexity." Authored by researchers at the University of Liverpool, the Chinese University of Hong Kong and other schools, the paper pointed out that achieving "human-like manipulation requires tightly integrated actuation, perception and control."  These attributes, however, increase mechanical complexity, computational demand and system cost, according to the research.

Dexterity seems to be on the rise, given videos of robots running, leaping and dancing. But while such demos generate enormous excitement, Patel said businesses buy the outcome and "that comes from a commercially ready product or offering."

John Moore is a freelance writer who has covered business and technology topics for 40 years. He focuses on enterprise IT strategy, AI adoption, data management and partner ecosystems.