Signals Inbox·July 16, 2026·Humanoid Robotics

Humanoids: what jobs can they do now?

Humanoids can already perform useful jobs today, but almost all dependable work is concentrated in factories and warehouses, where robots move standardized containers or place identical parts inside tightly controlled workflows.

Humanoid jobs ranked by current capability

Q1Which humanoid jobs are closest to being fully operational now?

The humanoid jobs closest to full operation today are short, repetitive transfers of standardized containers and rigid industrial parts. Across the strongest deployments, the winning workflow is remarkably consistent: the robot collects a predictable object, moves it through a controlled space and places it at a predefined destination.

The two best-documented programs have already produced more than 190,000 such transfers. Agility Robotics’ Digit passed 100,000 tote movements in a live GXO warehouse operation in November 2025. Figure 02 separately loaded more than 90,000 sheet-metal parts during 1,250 operating hours at BMW’s Spartanburg plant, contributing to the production of over 30,000 vehicles. No household, hospitality, healthcare, retail or construction task in our tracker comes close to that combined volume.

Even within factories and warehouses, the robots occupy a very specific part of the workflow. GXO did not ask Digit to search for products, prioritize orders or pack irregular purchases. It used the robot to transfer standardized totes between mobile robots and conveyors. BMW gave Figure identical parts and fixed assembly fixtures rather than a broad vehicle-assembly assignment. The productive unit is still one repeatable motion loop, not an entire human job.

The same choice keeps appearing elsewhere. Apptronik’s first commercial applications include case picking, palletization, workcell delivery and machine tending. Figure’s latest BMW workflow involves manipulating production materials and pulling a loaded cart. Boston Dynamics is preparing Atlas for industrial material handling and part sequencing, while Figure’s new agreement with Catalyst Brands starts in a distribution center. Four major developers have independently concentrated their early commercial work around the movement of goods between existing systems.

Developers are selecting tasks where the robot’s world can be made highly predictable without rebuilding the whole facility. Standard containers provide consistent shapes and grip points. Fixed destinations reduce perception errors. Warehouse and manufacturing software can decide when and where the robot should move. Humans deal with the weird cases.

That also explains why tote handling has advanced further than visually ordinary domestic chores. A towel or shirt changes shape after every contact. A tote remains almost identical across the first, thousandth and hundred-thousandth repetition. That difference is huge.

The closest operational humanoid job today is moving standardized totes, bins or rigid parts across short gaps between machines, carts and workstations. Humanoids can perform that narrow job repeatedly enough to create measurable customer value. Evidence for anything resembling a complete warehouse or factory occupation remains much weaker.

Q2Which humanoid tasks have actually moved from demos to real workplace pilots?

Material handling has produced nearly all the convincing transitions from humanoid demonstrations to sustained workplace use. The industry has demonstrated hundreds of skills, but customers repeatedly turn those broad capabilities into a much smaller set of assignments: tote transfer, part feeding, cart handling, case movement, inspection and machine tending.

Digit provides the cleanest progression. GXO began testing the robot in a live Spanx warehouse in late 2023. The companies converted that pilot into a multiyear Robots-as-a-Service agreement in June 2024. Digit then passed 100,000 tote movements in November 2025. By late 2025 and early 2026, Agility had also announced commercial agreements with Mercado Libre and Toyota Motor Manufacturing Canada. One pilot had become a repeatable commercial proposition offered across logistics and manufacturing customers.

Figure followed a different but equally revealing path. Figure 02 accumulated 1,250 operating hours at BMW and exposed hardware weaknesses that fed directly into Figure 03. The next-generation robot returned to Spartanburg in June 2026 for a more mobile logistics task involving a loaded cart and whole-body manipulation. The program did not simply repeat the original demonstration with newer hardware. It moved from fixed part placement toward a broader material-flow problem while retaining the same customer and plant.

Apptronik offers a third trajectory, although its operating evidence remains less quantified. Mercedes-Benz announced the first Apollo commercial pilot in March 2024. Apptronik later added relationships with GXO and Google DeepMind, while Mercedes-Benz remained an investor as the company raised more capital for production and deployment. The commercial focus has also narrowed into a recognizable set of industrial workflows rather than an open-ended promise that Apollo can perform any physical job.

Retail gives us a useful counterexample. Sanctuary AI reported that its robot completed 110 different tasks during a one-week pilot at a Mark’s store, including picking, packing, cleaning, tagging, labelling and folding. That pilot showed much greater task variety than Digit’s warehouse assignment, yet Sanctuary has not reported an equivalent multiyear store deployment or six-figure task volume. The broader demonstration was impressive. The narrow warehouse loop commercialized faster.

Task breadth has so far been a poor predictor of commercial progress. Repetition has been a much better one. A robot demonstrating 100 unrelated abilities may look more general, while a robot performing the same useful transfer 100,000 times is much closer to becoming a product.

The humanoid tasks that have genuinely left the demo stage are the ones a customer can isolate, measure and repeat thousands of times. Laundry folding, room tidying and complex retail handling continue to improve, but they have yet to produce comparable evidence of sustained customer use.

Q3Can humanoids work a full shift without human help today?

Humanoids can now participate in workflows scheduled across a full industrial shift, but no company has publicly shown that one can complete a varied eight-to-ten-hour job without human support.

Figure provides the strongest endurance claim. Its robots were scheduled for ten-hour weekday shifts at BMW and accumulated more than 1,250 operating hours, equivalent to roughly 125 ten-hour shifts. Yet the same deployment identified the forearm as the robot’s most frequent hardware failure point and generated substantial redesign work for Figure 03. The robots spent meaningful time inside production, but the published figures do not disclose how often people intervened, how long the robots remained continuously productive or how many failures occurred per shift.

Digit’s history gives us a different measure of endurance. The GXO deployment continued from a 2023 proof of concept through a 2024 commercial agreement and beyond 100,000 tote movements in 2025. Surviving that progression is stronger evidence than an isolated endurance test because the robot repeatedly returned to a live customer operation over roughly two years. Still, neither GXO nor Agility has published movements per productive hour, intervention frequency, recovery time or the proportion of warehouse operating hours actually covered by Digit.

Hardware road maps also show that shift-level operation is still being engineered. Figure redesigned components based on BMW failures. Boston Dynamics says Atlas can replace its own battery to continue operating. Agility now sells Digit with fleet software, support infrastructure and rapid on-site response instead of presenting the robot as something customers can simply switch on and leave alone. Those surrounding systems are part of the product.

The crucial missing metrics are surprisingly consistent across companies. Manufacturers publish total hours, task counts and scheduled shift lengths, but rarely reveal how many times a person rescued the robot, how much of the shift was productively spent moving goods, how often it stopped after an unsuccessful grasp, how long recovery took or how much support came from remote operators and on-site engineers.

Without those numbers, a clean comparison with human labor is impossible. A worker completing a shift manages blocked routes, damaged packaging, misplaced parts, station stoppages and dozens of minor exceptions without treating each one as a separate technical incident. Current humanoids perform best when another system or person absorbs those exceptions.

Humanoids can remain deployed within a full-shift operation today. They have not proved full-shift independence. The next meaningful milestone will be a disclosed shift containing uptime, interventions, successful cycles and recovery statistics, not another ten-hour headline.

Q4What will be the first humanoid job to reach real scale?

The first humanoid job to reach real scale will most likely be moving standardized totes, bins or cases between existing warehouse systems. We put the probability at roughly 70%, compared with 25% for repetitive automotive part handling and 5% for every other task combined.

Container transfer begins with the strongest commercial base. Digit has already passed 100,000 tote movements under a multiyear GXO agreement. Agility subsequently added deployment agreements with Mercado Libre and Toyota Motor Manufacturing Canada, giving it customers in third-party logistics, e-commerce fulfillment and automotive manufacturing. The underlying task can travel across those industries because standardized containers and internal material flows are common to all three.

Developer convergence points the same way. Apptronik targets case picking, palletization and workcell delivery. Figure has expanded from BMW part loading into cart movement and signed a logistics agreement with Catalyst Brands. Boston Dynamics is preparing Atlas for part sequencing, order fulfillment and machine tending. These companies use different hardware, software and commercial strategies, yet each has selected variations of the same basic material-flow problem.

Production capacity is also beginning to rise faster than the number of proven jobs. Figure says it manufactured more than 350 Figure 03 robots and increased production from one robot per day to one per hour in less than 120 days. Agility’s RoboFab has stated peak capacity of 10,000 robots annually, while Boston Dynamics has already committed all of its 2026 Atlas deployments. A growing hardware base paired with a short list of validated workflows makes replication more likely than sudden task diversification. The first fleets will probably repeat the few assignments that already work.

Automotive part handling remains the strongest rival. Figure’s 90,000 documented part placements are only about 10% below Digit’s 100,000-tote milestone. Automotive manufacturers also possess experienced automation teams, high labor costs and thousands of repetitive ergonomic tasks. But an automotive deployment is often tied to one component, fixture, model and plant layout. A standardized warehouse tote is easier to reproduce across unrelated customers.

Container transfer occupies an unusually suitable gap in existing automation. Warehouses already use mobile robots, conveyors, carts and putwalls. These systems frequently stop short of connecting physically because fixed automation would be expensive or inflexible. A humanoid can use the same aisles, container handles and transfer points designed for people, allowing the customer to automate the missing link without reconstructing the entire building.

There is one real caveat. Once a transfer workflow becomes stable and high-volume, a wheeled mobile manipulator, specialized gripper or fixed machine may eventually perform it more cheaply than a full humanoid. The human shape is most valuable while customers need flexibility across infrastructure originally designed for workers.

Even so, standardized container movement has the clearest path from one robot at one station to fleets working across several sites. It already has measurable throughput, repeat customers and strong cross-company convergence. No household, healthcare, construction or customer-service task currently comes close.