Anthropic’s robot index separates capability from affordability

Reader question: If robots can perform many physical tasks, why is near-term automation still limited?

Anthropic published a robot exposure index on September 30 that separates what robots can do from what employers can afford to automate. The research says robots can perform 74% of physical tasks in the United States, representing about 34% of working hours, but mostly in limited or structured settings. It estimates that robots are cost-competitive for just 0.3% of tasks.

That gap changes the business conversation. A capability map can show that an autonomous vehicle, warehouse system or hospital robot performs part of a job. It does not show that buying, operating and supervising the system costs less than paying people. Anthropic’s analysis says that if robot prices decline at historical rates, the share of tasks with cost-competitive robots would take about 40 years to reach 10%.

The index also names barriers beyond price. The research groups them as capabilities, human preferences and regulation. Manipulation and handling remain difficult for many tasks; people may reject a machine in work that depends on trust or contact; and rules can restrict automation in healthcare, education and protective services. Those constraints mean that a high exposure score is a signal for investigation, not a timetable for layoffs.

Anthropic says it used Claude to score detailed examples of job tasks, combined those judgments with employment and wage data, and checked the measure against historical changes. The resulting numbers are research estimates, not an independent census of every robot deployment. The source also notes that task exposure varies by setting: a machine that works in a warehouse may not work on an open street or in a home.

For a company deciding where to invest, the useful checklist is therefore capability, context, cost and permission. Ask what the machine can do today, where it can do it safely, what a complete deployment costs and which rules or preferences limit adoption. Anthropic’s index does not answer those questions for a particular employer, but it supplies a more realistic frame than treating technical possibility as immediate economic substitution.

Source: Can we predict the jobs robots will do?

This looks like a case where a smaller, well-defined decision may be more useful than a broad promise. What narrow use would you test conceptually before widening it?

A useful question for this decision is what evidence a team would require before expanding beyond the announced scope of “Claude identifies a CRISPR-like enzyme system, but its function remains open”. Which measure would you put first?

The trade-off here is between the capability described by the source and the governance needed around it. What review owner should sign off before this becomes a routine workflow?