Anthropic Finds Robotics and AI Reach 81% of Work Tasks, but Costs Limit Economic Viability to 0.3%

Combining artificial intelligence software and physical robotics could technically address roughly 81% of tasks performed by U.S. workers, according to new research from Anthropic reported by The Next Web (https://thenextweb.com/news/robots-and-ai-could-reach-80-of-work-but-rarely-cheaply-anthropic-says). While software-based AI models alone reach about half of all work, adding physical systems extends potential coverage into manual, lower-wage occupations.

Despite that broad technical ceiling, high hardware and operational costs sharply restrict current economic deployment. Anthropic found that robots are currently cheaper than human labor for only 0.3% of U.S. tasks, representing approximately 300,000 workers concentrated primarily in packing and packaging occupations.

The research illustrates how physical automation inverts the pattern typical of software models, which predominantly touch white-collar administrative and professional roles. In transportation, AI models alone expose less than 15% of tasks, but adding robotics raises total exposure to approximately 90%. Nine of the ten most exposed occupations in the sector are vehicle drivers, led by taxi drivers.

Technical feasibility also remains bound to structured environments. Anthropic calculated that robots can technically handle about 74% of physical tasks, though most implementations require controlled conditions such as factory assembly lines. In less predictable settings like cleaning and maintenance, automated equipment remains several times more expensive and slower than human workers.

Cost trajectories suggest a protracted transition. If robotics prices fell by 20%, the pool of cost-competitive roles would expand to about 2.8 million workers. However, based on historical rates of cost reduction, reaching cost parity across 10% of tasks would take approximately 40 years. Anthropic emphasized that task exposure does not automatically equate to job elimination, and generated its occupational ratings using its Claude model paired with web search.