A Job Description for the Room

A robot can do a surprising amount of physical work today, provided the room has already learned how to behave.

That condition sits inside Anthropic's new robot exposure index. The researchers sorted roughly 19,000 job tasks into four categories: work no current robot can perform, work possible in a purpose-built robotic environment, work possible in a structured human workplace, and work possible in an unstructured environment.

The resulting headline is large. Robots can perform 74% of physical tasks in the United States, representing 34% of working hours, in at least some circumstances.

The circumstances are doing quite a bit of work.

Nearly a quarter of all working time falls into the second category: tasks a robot can perform after a setting has been built for the robot. Another 10% belongs to structured human workplaces such as logistics warehouses. Only 1% of working time involves tasks robots can perform in unstructured places such as public roads.

There is a caveat before those numbers travel. Claude helped classify the task descriptions, estimate how workers divide their time, and model robot costs from cited examples. The researchers publish the task-level reasoning and test an earlier version of the measure against 50 years of labor outcomes, but this is still a proposed index of exposure, not a census of installed machines or displaced workers.

This is a useful taxonomy because it refuses to call every demonstration equally ready. It is also a reminder that capability does not live entirely inside the machine. Sometimes the machine becomes capable because the world around it has been simplified.

A package can be given a standard shape and a readable label. An aisle can be kept clear. A pallet can arrive at a fixed height. A person can be told where to stand. Work can be divided into motions a gripper repeats and exceptions somebody else resolves. The robot performs the task, but the room supplies part of the intelligence.

That environmental contribution tends to disappear from automation stories. We notice the arm, vehicle, camera, or model because it is the new object. We notice less often the floor markings, guarded zones, conveyor geometry, maintenance access, inventory discipline, and changed tempo that make the performance possible. None of those accommodations are incidental. They are an unwritten job description for the space and for the people still moving through it.

This has been visible for a long time. In 1985, the National Institute for Occupational Safety and Health warned in its guidance on preventing injuries from robots that replacing a person with a robot had to account for the robot's work envelope. A robot often needed more operational space than the person who had done the same task. The automation decision did not remove a body from a workstation and leave the room untouched. It redrew the room.

Contemporary warehouses make that redrawing easier to see. OSHA's current warehousing guidance names conveyors, labelers, pallet wrappers, and industrial robots as systems that can create struck-by and caught-between hazards when they are poorly integrated. The same guidance notes that fast work pace and continuous performance monitoring can intensify stress, fatigue, and musculoskeletal risk.

The machine does not need to hit someone to reorganize the body around it. It can determine where a person waits, how quickly they clear a lane, which variation counts as a defect, and how much time exists to notice that an ordinary day has become unsafe.

The new index does not ignore this complexity. Its exposure score gives less weight to tasks that require purpose-built settings, and its cost estimates include workplace redesign, maintenance, supervision, and repair. That discipline produces a much smaller economic claim: robots are currently cost-competitive for only 0.3% of work. Even cost parity is not the same as easy adoption. Remaining human tasks can become bottlenecks, and the estimates already depend on people supervising and handling exceptions.

Still, the phrase a robot can do this task is likely to travel farther than its environmental footnote. Capability numbers have a habit of reaching strategy decks without the rooms they came from.

Before treating a physical task as automatable, an organization should inventory what the setting must promise. Which objects need to become uniform? Which paths must remain empty? Which pace must stay predictable? What new inspection, cleaning, labeling, charging, guarding, or recovery work appears? Who handles the package that leans, the spill that changes the floor, the label that wrinkles, or the person who enters the aisle unexpectedly?

Dangerous and repetitive work does not become precious because automation is complicated. The NIOSH Center for Occupational Robotics Research points to real benefits: robots can inspect hazardous sites, reduce chemical exposure, and help with physically punishing lifts. Redesigning a workplace can make it safer. It can also make it more brittle, more surveilled, or less forgiving when the redesign serves throughput and treats human adaptability as free backup capacity.

That is the distinction worth preserving. A room can be structured for safety, clarity, and shared work. It can also be structured so the machine succeeds by asking people to absorb every irregularity the machine cannot.

The robot exposure index measures the difference between those rooms as degrees of technical readiness. A workplace decision has to ask one question more: what kind of place must this become before the robot's capability is true?