Most humanoid robots you read about exist in two states: impressive demonstration video, or speculative promise. Agility Robotics' Digit is different. As of late 2023, Digit units have been operating inside Amazon's fulfilment operations in the United States — not in a test environment, not as a one-off showcase, but in supervised production conditions, doing a specific job, repeatedly, on shift.
That fact deserves more careful attention than it's received. The coverage when Amazon and Agility announced their partnership in late 2023 was predictably breathless. The follow-up coverage — which should have asked harder questions about what "deployment" actually meant in practice — was largely absent.
Here's what we know, what we don't, and what it tells us about where humanoid robotics actually is right now.
What Digit Is
Digit is a bipedal humanoid robot built by Agility Robotics, a company founded in 2015 as a spinout from Oregon State University's Dynamic Robotics Laboratory. The robot stands roughly 175 centimetres tall and weighs around 65 kilograms. It walks on two legs, has two arms with gripper-style hands, and is designed to operate in environments built for humans — navigating aisles, handling bins, working around human co-workers.
The current generation of Digit uses a combination of cameras, lidar (a sensor that maps the robot's surroundings using laser pulses), and onboard computing to perceive its environment and execute tasks. Like most commercial humanoid deployments, Digit runs on a mix of autonomous operation for well-defined tasks and human oversight for edge cases.
Agility has been clear about what Digit is designed to do in its Amazon deployment: move empty tote bins from one location to another within fulfilment centres. That's the task. Not picking individual items, not packing boxes, not interacting with customers. Moving totes — a repetitive, physically demanding job that humans currently do and that requires just enough dexterity and mobility that a conventional wheeled robot can't easily do it.
What the Amazon Deployment Actually Involves
The gap between "robots deployed at Amazon" and what that means in practice is significant, and worth being specific about.
The Amazon-Agility partnership announced in late 2023 described a phased approach. Digit units were placed in Amazon facilities in a supervised testing context — not running autonomous full shifts unsupervised, but operating in controlled conditions with human oversight, with the task scope deliberately limited to tote handling. Amazon described the arrangement as evaluating Digit for potential broader use, not as a production rollout in the traditional sense.
What this means: Digit is real, the deployment is real, the task execution is real. But "deployed at Amazon" should not be read as "hundreds of Digit units autonomously working eight-hour warehouse shifts without supervision." The honest picture is closer to: a small number of units, operating a narrowly-defined task, in supervised conditions, as part of an evaluation programme.
That distinction matters because it tells us something true about where humanoid deployment actually is. The technology is far enough along to run structured pilots in real environments. It is not far enough along for unsupervised, general-purpose work at scale. Both things can be true — and often are, simultaneously, in the coverage of any given humanoid company.
Why Tote Handling Is the Right First Task
The choice of tote handling as Digit's initial deployment task is deliberate, and understanding why reveals something useful about how humanoid robotics will actually enter the workforce.
Tote handling has a specific set of properties that make it well-suited to early-stage humanoid deployment. The task is repetitive and well-defined — the robot isn't making decisions about what to pick or where to send it, just moving a standardised object between known locations. The environment is structured and predictable within the constraints of the task. The consequences of failure are low — a dropped tote is an inconvenience, not a safety incident. And the task is physically demanding enough that there's genuine operational value in automating it.
The progression from this kind of constrained, high-repetition task to more varied work is the real story to watch. Every additional capability — picking varied items, responding to unstructured environments, working safely next to humans in less controlled settings — adds complexity that current humanoid systems handle inconsistently. The tote task is a proof of concept for the physical deployment logistics. It is not a proof of concept for general humanoid utility.
The Economics Question No One Is Answering
The coverage of humanoid deployments consistently avoids the most important question: what does it cost per hour of reliable operation, and how does that compare to human labour?
Agility hasn't published detailed cost figures for Digit. Amazon hasn't commented on the economics of the programme. What we do know from industry estimates is that current-generation commercial humanoid robots cost somewhere between $70,000 and $200,000 per unit, with significant variation depending on capability level and production volume. Maintenance, support, and downtime costs add to the total cost of operation.
A human warehouse worker in the United States costs an employer roughly $35,000 to $55,000 per year in direct wages, plus benefits and overhead. A robot working 24 hours a day could theoretically provide more operational hours than a single human worker — but only if its uptime is high and its task success rate is reliable enough to justify the comparison.
The honest answer to the economics question right now is: we don't know. The data isn't public, the deployments are too recent and too limited to draw conclusions, and the companies involved have strong incentives to present the economics favourably. What we can say is that the economics will ultimately determine whether humanoid robotics expands into widespread deployment or remains a technically impressive but commercially marginal category.
What Agility's Competitors Are Actually Doing
Digit's warehouse deployment puts it meaningfully ahead of most humanoid competitors on the "actually deployed in production" metric — but the gap is narrower than the headlines suggest.
Boston Dynamics' Atlas, the most technically capable humanoid robot by most measures, has never been commercially deployed for production work. Atlas exists primarily as a research and demonstration platform; Boston Dynamics sells Spot, its quadruped robot, commercially, but Atlas is not a product you can buy. The company's recent shift to an all-electric Atlas represents a genuine technical advance, but production deployment remains unannounced.
Figure AI, which raised $675 million in early 2024 at a $2.6 billion valuation, announced a partnership with BMW to deploy humanoid robots in automobile manufacturing. Figure's demonstration videos — including a widely-shared clip of their robot handling objects at a table — are technically impressive. But the BMW deployment timeline and scope have not been specified publicly, and the company has not confirmed production-level autonomous operation. The honest assessment of Figure right now is: technically credible, commercially unproven, timeline uncertain.
Tesla's Optimus has received enormous coverage relative to its deployment status. Elon Musk has made repeated claims about Optimus deployment timelines — in Tesla factories, then for external sale — that have consistently been revised. As of early 2026, Optimus has been demonstrated performing factory tasks in Tesla's Fremont facility under supervised conditions. The company says units are working autonomously. Independent verification of that claim is not available.
The pattern across the industry is consistent: demonstrations are advancing, pilot deployments are real but narrow, production-scale autonomous deployment is largely future-state. Digit at Amazon is the most concrete example of the gap between demo and deployment beginning to close — and even there, the gap hasn't closed entirely.
What to Watch For Next
The meaningful signals to track in humanoid deployment over the next 12 to 18 months aren't the demonstration videos. They are: how many units are operating in which facilities, what tasks they're performing autonomously versus supervised, what the reported uptime and task success rates are, and whether any company publishes credible cost-per-hour figures that compare favourably to human labour at scale.
When those numbers exist and are independently verifiable, the humanoid deployment story will have moved from pilots to industrial reality. Digit at Amazon is the beginning of that story. It is not the middle, and it is not the end.