Anthropic Study: Robots Pay Off for 0.3% of Job Tasks

Anthropic says robots can already do three-quarters of physical job tasks but are cost-competitive with human labor for just 0.3% of tasks. Those are Anthropic’s own figures from a report of 30 September 2026 in which its model, Claude, scored the tasks and estimated the costs.

What Anthropic Published

Anthropic’s economics report “What work can robots do?”, dated 30 September 2026 and written by Russell Legate-Yang and Maxim Massenkoff, builds a robot exposure index. It says robots can already perform three-quarters of physical tasks in the US, which make up 34% of working hours, but mostly in limited settings.

It also says robots are cost-competitive for just 0.3% of job tasks. Anthropic builds the index by having its own model, Claude, score tasks and estimate costs, and the report says the rubric requires many judgment calls.

How the Index Is Built

The analysis uses O*NET, a database of around 900 occupations linked with descriptions of around 19,000 job tasks. Anthropic says it has Claude score task descriptions on a rubric measuring physical, cognitive and interpersonal requirements, then rate each physical task on four tiers.

The tiers run from E0, a robot cannot perform the task, through E1, a purpose-built robotic work environment such as a factory assembly line, and E2, a structured human work facility such as a logistics warehouse, to E3, an unstructured environment such as a city road.

Anthropic says it instructs Claude to search for specific robots relevant to each task, assess their capabilities and quote sources, that only demonstrated robot capabilities count, and that robots must do a task similarly well to humans, factoring in reliability, error rates and speed.

It says the O*NET task statements are often terse, so it elicits detailed examples of how tasks are done, and that exposure is set by majority rule: the least structured environment in which robots can do at least half of a task’s examples, weighted by time.

What It Finds on Exposure

Anthropic says driving and warehouse jobs are highly exposed to currently available robots, while nursing and general repair jobs are not, since present-day robots can do little of their work. It says exposed workers are more likely to be male, less educated and lower paid.

Counting both robots and large language models, the report says about 80% of job tasks by working time are exposed to one or the other. It says the remaining work is highly interpersonal or needs physical skills robots lack today.

Anthropic backtests the approach over 50 years. It says jobs more exposed to robots experienced greater declines in wages and employment than others, and that each year robots have become able to do about 2% of the physical work they previously could not.

The Cost Test

Anthropic calls a task exposed to cost-competitive robots when the robot’s annual cost is less than the labor cost, taken from the occupation’s total compensation scaled by the share of time spent on that task. It says Claude estimates the robot costs using web search, and that, given adoption frictions and cost uncertainties, the measure is “more suggestive.”

On that test the report says robots are cost-competitive for 0.3% of job tasks. It says that if robot price declines follow past trends, it will take 40 years for that share to reach 10%.

Its worked example is packers and packagers. Anthropic says the robots needed are estimated to cost over $2 million to purchase and install but replace the yearly work of around 14 workers, with fixed costs spread over a roughly 10-year life at an 8% cost of capital.

It says operating costs bring annual cost to around $45,000 to replace one human worker per year, against labor of around $49,000 with 97% of time spent on tasks robots can do, so robots cost about $2,500 less per year. This publication’s own arithmetic: 97% of $49,000 is about $47,530, and $47,530 less $45,000 is about $2,530.

For welders, Anthropic says the robots needed together cost around five times more than human welders. It says robotaxis are estimated to cost only around $7,000 more than taxi drivers but face regulatory hurdles. It says the cost estimates are approximate.

The report says the US employs around 560,000 packers and packagers, that employment in the occupation has fallen 22% since 2015, and that the Bureau of Labor Statistics projects it will shed the 11th-most jobs of any occupation by 2035.

What Anthropic Says It Means

Anthropic says that if the past is any guide, taxi drivers and warehouse packers will see changes sooner than nurses and mechanics, and that it expects physical work to be automated first where robots have a foothold today.

It also says AI-powered robots could leapfrog its scale and do work they cannot today, such as learning to climb ladders, and that it has not considered how AI could affect physical work without robots. Those are Anthropic’s views, from a company that sells AI models.

The Structural Read

The report’s four measures sit on different bases, so they should not be added or compared directly: 74% of physical tasks, 34% of working hours, about 80% of tasks by time exposed to robots or large language models, and 0.3% of job tasks where robots cost less than labor.

The cost test compares annual costs. Anthropic calls a task exposed to cost-competitive robots when the robot’s annual cost is less than the labor cost, and it says that, given adoption frictions and cost uncertainties, the measure is “more suggestive.”

Its worked example is packers and packagers. Anthropic estimates around $45,000 a year to replace one human worker against labor of around $49,000 with 97% of time spent on tasks robots can do, so about $2,500 less a year. The article’s own arithmetic puts the labor figure at about $47,530 and the gap at about $2,530.

Three Implications

POSSIBLE IS NOT THE SAME AS PAID FOR Anthropic says robots can already do three-quarters of physical tasks in the US, mostly in limited settings, and that robots are cost-competitive for just 0.3% of job tasks.

A MODEL SCORED THE TASKS Claude scored the tasks and estimated the costs, using web search for robot prices. Anthropic says the rubric requires many judgment calls and that the cost estimates are approximate, so the figures are model-assisted estimates, and this publication repeated none of the scoring.

TIMING IS STATED AS A CONDITIONAL Anthropic says that if robot price declines follow past trends, it will take 40 years for the cost-competitive share to reach 10%. This publication does not predict any job outcome.

What Is Not Established

This publication did not repeat any scoring or cost estimate and did not check the O*NET, BLS or robot price inputs. Because Claude produced the task scores and cost estimates, the figures are model-assisted estimates, and Anthropic itself calls the rubric judgment-heavy and the costs approximate.

All figures are Anthropic’s own and none was independently verified. Nothing here predicts any job outcome and nothing in this piece is investment advice.

Business Engineer Framework

The Map of AI — The Stack Beneath the Applications

The Map of AI places more than 200 companies across nine layers of the stack, from silicon to application. It shows the layers that sit beneath the applications people use every day.