Anthropic: robots can do 74% of physical tasks

Anthropic's economic research introduces a task-level robot exposure index that assesses thousands of tasks from the U.S. O*NET database. The study finds that modern autonomous machines can already execute roughly 74% of physical job tasks, representing about 34% of all U.S. working hours, primarily in structured environments like factory lines and warehouses. This technical capability does not translate directly into economic adoption, which remains constrained by cost, hardware dexterity, and regulatory hurdles.

The study also notes that robots are currently cost-competitive with human workers for just 0.3% of tasks. At historical hardware price-decline rates, reaching even a 10% cost-competitive threshold would take around 40 years. However, historical backtests spanning five decades confirm that occupations with higher initial task exposure reliably suffered relative declines in long-term wage and employment growth. This suggests that sectors already seeing robotic entry—such as commercial driving, packing, and logistics—will face economic disruption well before unstructured trades like plumbing or nursing.

The stakes hinge on whether hardware costs decline faster than the historical rate and whether regulatory hurdles ease. For workers in structured environments, the research points to a potential long-term shift in wage and employment growth, though the timeline remains uncertain given the current cost-competitiveness gap.