Agility Robotics combines Digit humanoids, Arc and industrial deployment

Agility Robotics develops Digit, a humanoid for industrial material handling, and Arc, the cloud software that coordinates its work. Its commercial history includes a customer-reported robots-as-a-service deployment with GXO. The newly announced Digit 5 is a separate readiness question: Agility explicitly labels it in development, with specifications and some safety features subject to change.

01 / ProductDigit performs the work; Arc connects the workflow

Agility’s solutions page combines three elements: Digit hardware, Arc orchestration and service support. Digit provides mobile manipulation, while Arc helps manage workflows, fleet state and connections to compatible autonomous mobile robots and warehouse systems. The software matters because moving an object is usually only one step in a process owned by several machines and people.

The Digit 5 product page describes the next generation, including adaptable grippers and a new approach to working near people. Its explicit development disclaimer governs those claims. A displayed payload, runtime target or safety feature is a specification under development, not a result that this article independently measured or a guarantee that a delivered installation will achieve it.

Agility’s September announcement connects the new design to experience with Digit 4 on customer floors. That history makes the company relevant to applied AI, but the generations should not be collapsed. The right vendor conversation identifies the hardware and software version being offered and which capabilities are available in that programme.

02 / AudienceA humanoid fits some gaps between existing automation

The deployment guide starts by asking whether a humanoid is appropriate at all. Repetitive movement between separate pieces of automation can be a plausible fit. Agility also says a fixed arm or autonomous mobile robot may be more suitable in some cases. This is a useful decision boundary for operators trying to solve a bottleneck rather than buy a particular shape of machine.

A distribution centre with goods arriving on mobile robots but needing transfer to a conveyor has a concrete starting point. GXO’s commercial announcement describes that sort of tote movement at a SPANX facility. The evidence establishes a named workflow and agreement; it does not establish that every warehouse uses the same containers or that a standard deployment requires no adaptation.

Teams with continually changing goods, fragile objects or ambiguous destinations need to examine their exception rate before assuming flexibility solves the problem. A humanoid that can be retasked still needs a defined correct outcome. Agility’s industrial focus is also different from a consumer helper: its public route is a facility deployment discussion, not a household robot subscription.

03 / WorkflowA proposed tote-transfer assessment

Consider a proposed evaluation of a tote handoff between a mobile robot and a conveyor. Sequenced has not tested Digit. Begin by recording the actual variations: tote sizes, weights, pickup positions, arrival timing and the conditions that prevent a clean transfer. Compare the robot proposal against the current process using the same boundaries, including waiting and recovery time.

Agility’s Customer Acceleration Program first defines a use case and validates it in controlled conditions at its own facility, using simulation and physical setups. That suggests an initial evidence packet built around representative objects and layouts. A customer should ask which details can be reproduced before equipment reaches its site and which only emerge during live operation.

On-site validation then needs to include upstream and downstream coordination. A transfer can be mechanically successful while still causing the wrong order to enter the next stage. In this proposed test, every job would retain its identifier, expected destination and final disposition. An interruption should leave an understandable state so staff know whether the goods still need to move or the software merely needs confirmation.

Record completed handoffs, unsuccessful attempts, human assistance and time spent waiting for other equipment. Keep safety stops distinct from task failures; both affect throughput but require different investigation. Site-qualified personnel and the vendor must determine the approved operating arrangement. This article does not provide a control configuration or treat a product safety claim as permission to remove physical safeguards.

The final output should be a narrow acceptance decision with known exclusions. If the robot handles a stable tote family but not distorted containers, that can still be a useful deployment boundary. The financial case should include the people who manage the excluded work and recover exceptions, rather than assuming all of their time disappears.

04 / PricingRobots-as-a-service is documented; a universal tariff is not

GXO describes a multi-year robots-as-a-service agreement covering Digit and Arc. This supports a commercial service model and a real customer route. It does not publish the payment schedule, included hours or a rate that another customer can claim. A quoted industry cost per hour is not a current Agility price unless it appears in that buyer’s offer.

The current sales page invites a deployment conversation. Ask which commercial structure is offered for the relevant generation and whether it covers task adaptation, integration, maintenance, support and software updates. Those are questions about the proposal, not assumptions that every agreement uses the same bundle. Digit 5 plans need a delivery and acceptance definition as well as a number.

Budget against usable throughput in the planned shift pattern. A monthly fee can be attractive or expensive depending on how much completed work the system contributes and how much time people spend supporting it. Do not convert a maximum runtime or vendor target into billable productive hours without testing the corresponding job.

05 / DistinctionsDeployment experience is as relevant as physical capability

Agility’s combination of a defined adoption programme and fleet orchestration distinguishes it from a robot research demonstration alone. The solutions description includes on-site service, online support and monitoring. These features matter when a robot must be integrated into an operating process with existing equipment, planned shifts and accountable support staff.

The Figure blueprint is a useful humanoid comparison for industrial learning and manufacturing. Agility’s published material offers a more explicit description of the stages from initial use-case screening to site validation. This does not prove one vendor performs better; it gives a buyer a concrete framework for comparing how a project will be delivered.

The NVIDIA blueprint provides context for the infrastructure used across physical AI. Agility’s customer is evaluating a supported system that moves real goods, rather than selecting compute or downloading a model. Capability, integration and operating responsibility therefore belong in the same evaluation.

06 / QuestionsDigit 5 development claims need their own acceptance evidence

The central near-term question is what can actually be delivered. The Digit 5 page says features and specifications may change and some safety features remain in development. Keep that caveat visible when reading the more ambitious announcement language about working close to people. An organisation needs the assessment applicable to its installed version and environment.

A second question is how new tasks are introduced after deployment. Cloud-delivered capabilities may reduce some adaptation work, but they do not establish that an unfamiliar workflow can be approved without testing. Ask what evidence accompanies a changed gripper, object family or task sequence, and who can authorise the transition in the customer’s operation.

Finally, clarify responsibility when the robot and other automation disagree about a job’s state. A meaningful service arrangement covers diagnosis across that boundary, including access to useful event records. Otherwise, a team may find itself resolving disputes between systems while the physical work queues up.

07 / DecisionBuy an accepted workflow, not an implied future capability

Agility merits attention for connecting humanoid hardware with orchestration, support and documented commercial work. The strongest next step is a well-scoped facility discussion grounded in the existing process. Readers following Digit 5 should distinguish what the new generation promises from what has already been validated and offered for their use.

The decision becomes clearer when a business can name the goods, transfer points, expected pace and exceptions. That information allows a comparison with conventional automation as well as other humanoids. A pilot is valuable when it narrows uncertainty about a real bottleneck, even if it concludes that a simpler mechanism fits better.