What work can robots do? news update: What work can robots do?

Anthropic released an empirical research study detailing a new robot exposure index to evaluate how autonomous physical systems intersect with the American labor economy. The investigation reveals that autonomous machines can already perform 74 percent of physical job tasks in the United States, representing 34 percent of total working hours. As documented in Anthropic Research's report, researchers analyzed thousands of occupational benchmarks to measure mechanical capabilities, environmental constraints, and capital costs across the contemporary industrial workforce.

This report examines the technical boundaries, environmental classification tiers, demographic disparities, and capital barriers shaping physical automation across commercial industries. The analysis is structured for engineering directors, roboticists, operations managers, and technology executives evaluating enterprise automation timelines. It provides an objective assessment of where autonomous equipment can replace manual labor today and explains why broad adoption across unstructured environments will take decades.

01Physical task coverage across working hours

Autonomous machines currently possess the mechanical capabilities needed to complete 74 percent of physical work tasks in the United States, representing 34 percent of total working hours. These systems operate primarily within tightly controlled environments rather than dynamic human spaces. Unexposed tasks remain insulated because they require delicate physical dexterity or interpersonal communication that contemporary machines cannot supply.

When combined with software automation from large language models, approximately 80 percent of all job tasks by working time face technological exposure. Physical machines address specialized manual functions where language models cannot execute real-world actions. The remaining one-fifth of unexposed employment consists of roles that demand human empathy, relationship building, or fine physical manipulation beyond current mechanical abilities.

02Four environmental tiers of task exposure

Researchers measured automation exposure across roughly 900 occupations and 19,000 tasks from the federal O*NET database by evaluating the environmental control required for operation. To evaluate each role, Claude analyzed task descriptions across physical, cognitive, and interpersonal requirements using documented real-world deployments. Demonstrated mechanical abilities must match human speed, error rates, and operational reliability to qualify as viable task exposure.

Task exposure ratings rely on a majority rule standard where robots must execute at least half of a task's real-world examples weighted by time. For example, autonomous taxis advanced from purpose-built test facilities in the early 2010s to unstructured public streets today. However, machines continue to struggle with variable manual tasks like untangling wiring or navigating cluttered, unpredictable job sites.

03Capital expenditure and forty year cost parity

While automated machinery can physically execute most manual tasks, current commercial hardware remains substantially more expensive than human labor. Autonomous systems are cost-competitive for only 0.3 percent of job tasks across the national economy today. If machine price declines follow historical trajectories, cost-competitive automation will require 40 years to reach just 10 percent of domestic employment tasks.

Capital expenses are not the sole factor restricting the spread of autonomous systems across commercial workplaces. Significant technical hurdles persist, including the complex physical dexterity required for fine assembly and object manipulation. Legal frameworks, liability risks, regulatory mandates, and cultural preferences for human service workers present additional institutional barriers that delay automated equipment purchases.

04Historical wage declines across exposed occupations

A historical backtest across 50 years of labor data validates that jobs with high robot exposure experienced sustained economic declines. Between 1977 and the present, occupations facing direct mechanical exposure suffered greater drops in relative wages and total employment. Over that same half-century window, autonomous machines expanded their reach, absorbing roughly 2 percent of previously unautomated physical work each year.

The labor impact falls unevenly across demographic groups in the current workforce. Workers in highly exposed occupations are more likely to be male, possess lower levels of formal education, and receive lower median compensation. Transportation drivers and warehouse workers face immediate exposure risks, while healthcare workers and skilled mechanics remain shielded due to unstructured working conditions.

05Industrial adoption rates and humanoid factory testing

Annual enterprise expenditure on robotic hardware accounts for approximately 1 percent of total equipment investment across the United States. Despite this modest baseline, business surveys suggest domestic commercial adoption could nearly double over the next three years. Venture-backed engineering companies are investing billions of dollars to build general-purpose humanoid robots capable of matching human physical capabilities.

Initial testing of AI-powered humanoid platforms remains concentrated inside automobile manufacturing plants and logistics warehouses where automated machinery is already standard. Deploying in structured industrial environments allows manufacturers to refine machine perception alongside human workers before addressing unpredictable consumer spaces. Physical work will experience automated displacement first in controlled facilities where robotic platforms already possess an established operational foothold.

Note: Commercial viability remains strictly constrained by non-price barriers such as fine motor dexterity, regulatory safety mandates, and strong customer preferences for human workers. Even when mechanical hardware matches human speed and reliability, these institutional requirements keep automated deployments confined to structured industrial environments for the foreseeable future.