Anthropic maps the gap between robot capability and affordable automation

What the research finds

Anthropic’s September 30 study estimates that robots could perform tasks accounting for 34% of US working time under some conditions. Its modeled costs make automation competitive with human labor for just 0.3% of working time. Claude assesses robot evidence against occupational tasks and estimates time spent on them; employment weights come from 2025 BLS data. Much of the technical exposure requires controlled environments. These are estimates of potential task coverage and economics, not measured deployment or job losses.

Sources: Anthropic

A consequential methodological limit

The appendix says some ratings infer capability from robots doing related tasks. Excluding these ratings reduces exposed physical work from roughly three-quarters to half. Historical labor regressions pool three overlapping periods spanning 1977–2024. Their latest labor observations use the 2020–2024 American Community Survey; those observations should not be mistaken for September 2026 employment outcomes. The historical comparisons support an association, not a fresh count of workers displaced by AI.

Sources: Anthropic

Our interpretation: inspect the whole shift

For a worker, a useful starting point is to write down what happens during an ordinary shift and during a difficult one. Which activities repeat predictably? Where do colleagues step in? What happens when equipment, supplies or customer requests differ from the usual pattern? This creates a concrete basis for discussing a proposed installation with a manager or worker representative.

We would ask employers to describe the intended staffing decision separately from the equipment demonstration. Will a pilot change output targets, overtime, recruitment or existing positions? Who will receive training, and when? A demonstration can help answer a technical question while leaving these organizational decisions unresolved. Workers need both kinds of information to plan. A written review date would also give everyone a point at which to compare the original proposal with what the pilot actually delivered.

Our practical next step

Keep a short record of announced pilots and their actual outcomes in your workplace. Record what was promised, what work changed, and which responsibilities people gained or lost. Use that evidence when discussing training or a move into another role. Avoid paying for a generic robotics credential solely because a national exposure estimate sounds alarming; first identify a specific opportunity and the skills its employer asks for.