Robots are technically able to do roughly three-quarters of the physical tasks in the U.S. economy but are cheaper than human workers for only a sliver of them, according to research Anthropic published Wednesday. The study, by Anthropic researchers Russell Legate-Yang and Maxim Massenkoff, builds a robot exposure index that extends the company's earlier work on how large language models touch different jobs.

What Happened

The researchers used Claude to rate today's robot capabilities against about 19,000 tasks across some 900 occupations in the U.S. Labor Department's O*NET database. Each task was placed on a four-step scale. E0 means robots cannot do it. E1 means they can, but only in environments built around them, such as assembly lines. E2 covers structured human workplaces, and E3 covers unstructured settings such as city roads. Ratings required evidence of commercial sales, deployments or verified demonstrations.

By that measure, robots can perform 74% of physical tasks, which make up 34% of U.S. working hours. Most of that capability is confined to controlled settings: about half of physical tasks fall in E1 and another 22% in E2. Only 2% reach E3. Tasks robots cannot do at all account for about a quarter of physical work time.

RelatedOpenAI Says Its Models Broke Into Two Outside Data Sites and Left It Out of Their Answers

Cost changes the picture sharply. The authors estimate robots are cost-competitive for just 0.3% of job tasks. Robot prices have historically fallen about 3% a year, and at that rate the cost-competitive share would take about 40 years to reach 10%.

Why It Matters

The index separates what robots can do from what is worth paying for. Packers and packagers are the largest occupation exposed to robots that already make economic sense: systems costing more than $2 million to buy and install can replace the annual work of about 14 people, at roughly $45,000 per worker-year in operating cost against about $49,000 in wages. Welding is the counterexample, where the robots needed would cost around five times as much as human welders.

Exposure is concentrated in vehicle operation. Nine of the ten most exposed occupations are vehicle operators, and taxi drivers rank first with an index score of 2.2 out of 3. Workers in the most exposed jobs are 55 percentage points less likely to hold a bachelor's degree, earn around $30 less per hour and face an unemployment rate more than twice as high as unexposed workers, the study found.

Taken together with Anthropic's earlier LLM measurements, about 80% of job tasks by working time are now exposed to either robots or language models.

How the Method Was Tested

The authors backtested their approach over roughly 50 years and found that, from 1977 on, jobs more exposed to the robots of the time saw larger wage and employment declines in later decades. They also estimate that robots gain the ability to do about 2% of previously impossible physical tasks each year. Manipulation is the largest gap: half of physical tasks would not be automated at scale unless robots get better at touching and handling objects, the paper says.

What to Watch

In a fast scenario, where quality-adjusted robot costs fall up to four times faster and new capabilities arrive twice as quickly, robots become cost-competitive for half of physical work by 2050, the authors estimate, against 2085 at the historical pace. Anthropic plans to update the index periodically, and drivers and warehouse packers are the groups where early signs of displacement would show up first.