One hundred thousand totes gave Digit a useful work record

2025 archive · Agility’s November 2025 milestone documented repeated warehouse handling at GXO, while leaving the full economics of humanoid labor undisclosed.

Share LinkedIn X Facebook
Robots in an Ocado warehouse, 2017. File photo of a different automation system.
Robots in an Ocado warehouse, 2017. File photo of a different automation system. Photo: Techwords · CC BY-SA 4.0 · resized

Moving one tote makes a demonstration. Moving more than 100,000 creates an operating history worth examining. On November 20, 2025, Agility Robotics reported that its Digit humanoid had passed that cumulative count at GXO’s Flowery Branch facility in Georgia.

The figure was a vendor-reported milestone at an identified commercial site. It gave the emerging humanoid industry a measure of repeated work, although it did not by itself show cost per movement, uptime or the number of robots and operating hours needed to reach the total.

The deployment had an established commercial foundation. GXO and Agility announced a multiyear robots-as-a-service agreement in June 2024 following a late-2023 pilot. At the SPANX facility, the task connected existing automation: Digit moved totes from other mobile robots onto conveyors. Agility’s Arc software coordinated the work and supported activities such as mapping, workflow definition and troubleshooting.

That setting was more informative than a robot performing an isolated maneuver. A transfer between two pieces of equipment has to fit the pace and physical arrangement of the rest of the operation. The humanoid was useful only if the surrounding workflow could depend on its contribution.

Agility’s November account said Digit had also stacked totes at a separate floor location. The company presented that additional behavior as evidence of versatility. The reported examples remained closely related material-handling tasks, but they showed why the company was pursuing a mobile machine with arms: the same basic form could move between pickup and placement locations.

The technical account emphasized the repetitive difficulties of warehouse work, including balancing with different loads and identifying and grasping objects under varying lighting and placement conditions. Those challenges become consequential over many cycles. An occasional failure that seems minor in a short demonstration can interrupt an operation repeatedly across a shift.

A cumulative count is useful because it establishes that more than a handful of successful cycles occurred. It is incomplete because it lacks a denominator. Without the corresponding operating time, human-assistance requirements and service costs, readers cannot calculate a throughput rate or a return on investment from the 100,000 figure.

The commercial agreement also changed the purchasing question. GXO described buying the capability through a service arrangement, with fleet-management software included, rather than announcing only an equipment purchase. That made continuing operational performance central to the relationship. The publicly inspected records did not disclose the contract’s pricing or performance guarantees.

Humanoid Global, an investor in Agility, repeated the milestone in a November 27 filing and identified November 20 as the announcement date. That corroborated the chronology, not the underlying tote count through an independent audit. The count remained Agility’s report.

For warehouse operators considering the category, the useful development was the accumulation of ordinary work at a customer site. It provided a more concrete reference point than a planned factory or a promise of general capability. The next comparisons would need to be equally practical: completed transfers per paid operating hour, interventions, availability and the cost of fitting the robot into the job.

Digit’s six-figure total was a substantial entry in that record. It was a reason to ask more precise commercial questions, with an actual deployment to ask them about.

Sources