What the two companies put into figures
Daniel Diez, chief business officer at Agility Robotics, said the humanoid Digit v5 will start going in with early customers in December, can work beside people without a safety cage and can put in 20 hours a day. The sentence in which the company sets out its unit economics is the most concrete part of the announcement: on Diez's account the maths gets interesting at the point where one robot covers 3 shifts. That prices the robot against a work schedule rather than against a task.[1]
Boston Dynamics disclosed a duration the same day. The production version of Atlas recognises on its own that its battery is running down, walks to a designated station, replaces the battery in under 3 minutes and returns to its task. On the company's figures a conventional recharge takes about 90 minutes, while the robot runs for roughly 4 hours on ordinary work and about 2 hours under heavy load. What the announcement measures is shorter downtime.[2]
The measure missing from both
Both companies quantified the same thing: how long the machine can stay up. The human-side measures that belong opposite that figure are absent from both announcements. Working without a cage means the robot enters a space where people are; how many of them remain per shift, who resolves the exceptions the robot cannot, and which staff carry out safety validation are not written down. Concealment need not be the only explanation: the numbers may not exist yet because deployment has not begun, or they may be withheld as commercially sensitive, and either produces the same gap. Whatever its cause, the productivity claim on the table today can be measured only on the machine's side.[1], [2]
This moves one step past the question I left open in the August 4 column in Eigen Radar on the share of the worker who trains the robot. There the unsettled part was the share going to the worker doing the training; here it is the unit in which the gain is measured. A gain measured in hours does not convert by itself into a cost measured in people. A robot staying up across 3 shifts is compatible with the headcount on those 3 shifts falling, rising or staying put, and only the operator's own data can say which happened.[3], [1]
What should we watch for in December?
Because Agility says it will install with early customers in December, there is a concrete threshold on which the claim can be tested. If one of the customers announcing an installation publishes, alongside the robot count, the headcount per shift, total hours worked or contractor use at the affected operation, then whose ledger the gain lands in becomes measurable. Without such a disclosure we are left with the machine's uptime alone, and the distribution question stays open.[1]