Anthropic’s new robot-exposure index estimates that current robots can perform tasks accounting for 74% of U.S. physical work time, or 34% of all work time. But robots are cost-competitive with human labor for only about 0.3% of work, the company’s researchers report.
The index asks whether a robot can perform each task and what kind of setting it needs. Anthropic used Claude to assess 7,594 physical tasks drawn from O*NET descriptions covering roughly 900 occupations, and checked cited robot demonstrations, sales or deployments. A task receives more exposure when robots can do it in less controlled surroundings, ranging from a purpose-built production line to a warehouse or an open road.
That setting matters. About half of physical work time falls into tasks robots can do only in purpose-built environments. A further 22% of physical work time involves tasks robots can handle in structured workplaces, while 2% involves tasks they can perform in unstructured environments. Driving and warehouse work rank high in the index; nursing and general repair rank low.
The report’s barrier analysis identifies capability limits for about 70% of physical tasks. Manipulation—touching and handling objects—is the main missing skill for roughly half of physical tasks, while planning and reasoning are cited as a barrier for 8% of physical tasks. These are Claude-assisted assessments based on the researchers’ task evidence, not direct measurements of robots completing every job task.
Economic viability narrows the picture further. Anthropic estimates the full cost of deploying robots for a task and compares it with labor costs. If robot prices continued falling by about 3% a year, its scenario puts cost parity with human work at 10% of tasks about 40 years from now. The estimate holds tasks and wages fixed, so it is a conditional projection rather than a timetable for job replacement.
In a separate historical backtest, the researchers reconstructed robot exposure at several points since 1977 and compared it with later wage and employment changes. Jobs more exposed to the robots of their time later saw larger declines, even after accounting for industry trends and other factors. That association does not show that robots caused those changes. The index is designed to track where today’s robots have a foothold, not to predict when workers will be displaced.
Sources
- Anthropic: “What work can robots do?”
- U.S. Bureau of Labor Statistics: May 2025 OEWS tables
- O*NET database
Image credit: Rlistmedia, “Autonomous Mobile Robot AMR”, via Wikimedia Commons, licensed under CC BY 4.0. The image is illustrative and is not part of the Anthropic study.
