Anthropic has published a study estimating which US job tasks today’s robots can perform, and where they can do them. Its central finding is a useful brake on automation headlines: robots can perform an estimated 74% of physical tasks, but are cost-competitive with people for just 0.3% of work.
Anthropic Watch analysis
What happened
The study defines robots as autonomous physical machines that sense and act, then assesses tasks by how controlled an environment a robot needs. Anthropic says robots can do tasks amounting to 34% of working hours, mostly in structured settings such as factories and warehouses. Only 2% of physical tasks are rated possible in unstructured environments.
The researchers used Claude to assess tasks from the US ONET occupational database, drawing on cited examples of robot deployments, sales and demonstrations. They also compared their measure with historical job data, finding that jobs more exposed to robots from 1977 onwards saw greater wage and employment declines in later decades.
Why it matters
The study offers a way to discuss robotics and work that separates what machines can do from what businesses can afford to automate. Anthropic estimates that, if robot prices fall at past rates, the share of work for which robots are cost-competitive would take 40 years to reach 10%. Capability, workplace conditions, worker preferences and regulation could also slow adoption.
The historical comparison is a signal, not a timetable for any particular job. The study measures task exposure, and its authors acknowledge judgement calls in turning short job descriptions and varied work settings into ratings.
Our read
This is a more useful map than another sweeping forecast of which jobs robots will “take”. Driving and warehouse tasks look more exposed today than nursing or general repair, according to Anthropic’s ratings. But exposure is not the same as a robot arriving on Monday with a clipboard and a redundancy plan: cost and working conditions still matter.
What to watch
- Whether Anthropic’s accompanying task-level data and cited robot examples stand up to scrutiny.
- Whether robot prices and capabilities change the study’s cost and exposure estimates.
- How employers use task-level measures when making decisions about investment and work.
Discussion spark: Should debate about robot-driven job change focus on what machines can do today, or on how quickly falling costs could make those capabilities commercially attractive?
Sources and evidence
- Can we predict the jobs robots will do? – Anthropic (30 September 2026, 16:01 UTC)
Anthropic Watch is independently operated by WittyWires. It is not affiliated with, endorsed by, or operated by Anthropic.