Anthropic’s September 30, 2026 study estimates that robots can perform 74% of physical job tasks in the U.S. in at least some settings. Those tasks account for 34% of all U.S. working hours, while the study puts the share of work for which robots are currently cost-competitive at 0.3%.
What Anthropic’s study estimates
The study examines U.S. occupations and employment data. Its 74% figure describes physical tasks that robots could perform in at least one setting; those tasks represent 34% of working hours across the economy. The figures measure task exposure, not work already transferred to robots.
Cost competitiveness is a separate calculation: Anthropic compares modeled robot costs with labor costs. On that measure, robots are cost-competitive for 0.3% of work today. Being able to perform a task, in other words, does not by itself make automating it economical.
Task capability is not job replacement
Anthropic defines a robot as an autonomous physical machine that senses and acts. The study rates tasks on an E0–E3 scale, from tasks current robots cannot perform to tasks they can perform in unstructured environments. Its exposure figure counts physical tasks robots can perform in at least one of the settings rated E1 through E3.
The 74% and 34% figures therefore describe estimated technical capability and the share of working time tied to those tasks. They are not counts of jobs eliminated or robots deployed.
| Measure | Estimate or scenario | What it represents |
| Physical-task exposure | 74% of physical tasks; 34% of all U.S. working hours | Tasks robots can perform in at least one setting, weighted by working time and employment. |
| Cost competitiveness today | 0.3% of work | Work for which modeled robot costs are below labor costs. |
| Baseline threshold | About 40 years to reach 10% of human work | A scenario based on robot prices declining about 3% per year. |
| Baseline physical-work threshold | 2085 for half of physical work | A separate baseline scenario for cost competitiveness. |
| Fast scenario | 2050 for half of today’s physical work | Assumes costs decline up to four times faster and capabilities grow twice as quickly. |
What the cost scenarios mean
The roughly 40-year estimate uses an assumption that robot prices continue to fall by about 3% annually, a pace the study says has held since the 1990s. Under that scenario, the threshold is cost competitiveness for 10% of human work. A separate baseline scenario puts the point when robots become cost-competitive for half of physical work in 2085.
The 2050 result comes from a faster scenario, not the same threshold under the same assumptions. It assumes robot costs fall up to four times faster and capabilities improve twice as quickly, reaching cost competitiveness for half of today’s physical work. These are modeled outcomes, not dates when the study predicts jobs will disappear.
How the study builds its estimates
Anthropic used O*NET descriptions covering about 19,000 tasks across roughly 900 occupations, then rated 7,594 physical tasks. Claude-assisted examples and capability assessments drew on robot deployments, commercial sales, and demonstrations. A task’s rating reflects the least structured setting in which robots could perform at least half of its examples, weighted by time.
Combining task-level cost estimates into whole occupations brings its own complications. Anthropic warns that aggregation can count the same robot more than once or miss coordination costs. The study’s estimates describe U.S. work, with those modeling choices shaping both its exposure figures and its cost scenarios.
