Robots may already have the technical capability to perform nearly three-quarters of the physical tasks workers do in the U.S. But a new analysis from AI company Anthropic suggests that widespread workplace automation remains a long way off, with significant implications for how HR leaders approach workforce planning, job redesign, and employee development.
The research paper, titled What Work Can Robots Do?, found that robots could perform 74% of physical job tasks in the U.S., representing approximately 34% of all human working hours. Yet robots are currently cost-competitive with human workers for just 0.3% of tasks, equivalent to roughly 300,000 jobs.
For HR executives and business leaders, the findings highlight an important gap between what technology can accomplish and what employers can realistically afford to automate. They also underscore the need to prepare employees for changing job responsibilities rather than assuming technological capabilities will translate directly into widespread job elimination.
“Just because a robot can do the work doesn’t mean it should — or will,” said SHRM Chief Knowledge Officer Alex Alonso, SHRM-SCP. “The distinction between technical capability and economic viability is everything. We’ve become enamored with what technology can accomplish while paying far too little attention to what makes economic sense.”
Alonso said the disparity between robots’ technical capabilities and their current economics should inform how employers evaluate automation opportunities.
“Technological possibility is not the same as economic inevitability,” he said. “Employers don’t adopt technology simply because it works. They adopt it because it creates value; but in the case of AI that value is only realized with a higher degree of human capital investment via training and better responsible governance protocols.”
Where Robots Are Most Likely to Make Inroads
Anthropic’s analysis examined approximately 19,000 task descriptions across nearly 900 occupations using the U.S. Department of Labor’s O*NET occupational database. The researchers used Claude to evaluate the physical, cognitive, and interpersonal demands of each task, then searched for evidence that robots could perform those activities in real-world settings.
The analysis classified robot capabilities according to the environments in which machines could operate. Robots capable of working in purpose-built settings, such as factory assembly lines, can perform nearly half of physical tasks. In structured workplaces such as warehouses, they can perform approximately 22% of tasks. Their capabilities are much more limited in unstructured environments, such as construction sites and city streets. Only about 2% of all tasks fall into the category of work robots can perform in unstructured environments.
Driving-related occupations are among the most exposed, the research found. Taxi drivers ranked highest in the analysis, with an exposure score of 2.2 on a scale of zero to three. Other potentially exposed occupations include industrial truck and tractor operators, agricultural equipment operators, mining equipment operators, and shuttle drivers.
Warehouse packers and packagers also face relatively high exposure because robots can already perform activities such as moving materials, inspecting goods, assembling boxes, applying labels, and sealing packages.
The economics are particularly compelling in some warehouse settings. Anthropic estimated that a robot capable of performing work equivalent to 14 packers could cost less than employing those workers over the robot’s expected 10-year life, even after accounting for a $2 million purchase price and $45,000 in annual operating costs.
Still, exposure does not mean an entire occupation will disappear. The research found that automation vulnerability can vary considerably within a single job. Some drywall finishing activities, for example, are relatively easy to automate, while initial taping work presents challenges because the materials can damage robot components.
Jobs Will Change Before They Disappear
Alonso cautioned against interpreting task-level automation estimates as predictions of equivalent workforce displacement.
“We need to stop confusing the displacement of tasks with the displacement of workers,” he said. “Jobs are bundles of tasks, not individual activities. Automating one activity doesn’t necessarily eliminate a job; it often changes the composition of that job and the value of the person performing it.”
Instead of focusing exclusively on how many tasks machines can perform, HR leaders should consider how automation will change the skills and capabilities employees need to contribute.
“That’s where I believe the next great productivity opportunity lies — in redesigning work around the complementary strengths of humans and machines, rather than treating them as interchangeable,” Alonso said.
That approach requires employers to examine jobs at the task level, identify activities that could be automated, and determine where human judgment, problem-solving, interpersonal skills, and specialized expertise will remain essential.
It also places greater responsibility on HR to work with business leaders on workforce planning, reskilling, and organizational design. Employees whose jobs are significantly affected by automation may need training to assume new responsibilities, operate alongside robotic systems, or move into different roles.
A Longer Timeline Doesn’t Mean Less Urgency
Anthropic’s research suggests that cost, rather than technical capability, is currently the biggest obstacle to broader adoption of robotic automation. The report estimates that robot costs would need to decline approximately 70% for the share of tasks where robots are cheaper than human workers to reach 10%. Based on historical price declines of roughly 3% annually, that milestone could take approximately 40 years.
However, Alonso warned HR leaders against treating that projection as a reason to delay workforce preparation.
“The 40-year projection should be a wake-up call for workforce strategy, not an invitation to complacency,” he said. “I wouldn’t interpret this research as evidence that employers have four decades to prepare. Technology rarely advances in a perfectly linear fashion, and economics can change dramatically when innovation, scale, and labor shortages converge.”
He added that disruption will not be distributed evenly across occupations and industries. Some employers could face substantial changes much sooner, particularly in industries where robots already have a foothold and the economics of automation are becoming more favorable.
“The greatest risk isn’t that robots replace everyone tomorrow,” he said. “It’s that organizations wait until the economics become irresistible before they begin preparing their people for a fundamentally different world of work.”
Was this resource helpful?