Every week there is a new video of a humanoid robot folding laundry, carrying boxes or doing a backflip. If you run a trades business, a warehouse or a field-service company, you may be wondering whether any of this is coming for your work, and when.


This post separates three things: what "physical AI" means, what is documented in real workplaces as of November 2026, and what is still a demo or a promise. We rely on company statements where that is all there is, and we say so when it is.


What "physical AI" means


"Physical AI" is a label, not a standard. NVIDIA, which uses the term heavily, describes it as AI systems that let machines such as robots and autonomous vehicles perceive, understand and act in the physical world. In plain words: software that takes in camera and sensor data and decides what a machine should do with its arms, wheels or steering.


It is different from a chatbot in one important way. A chatbot's mistake is a wrong sentence you can read before you use it. A physical system's mistake is a dropped part, a collision or an injury, and it happens in a fraction of a second. That is why physical AI moves more slowly from demonstration to daily use.


Most physical AI in workplaces today is not humanoid. The International Federation of Robotics (IFR) reported 542,000 industrial robots installed worldwide in 2024, and 4.66 million in operation, in its World Robotics 2025 report. Those are mostly fixed arms and other conventional machines in factories. The page announcing the figures does not mention humanoids.


Demos versus deployments


A demo shows a robot can do a task once, under chosen conditions, usually filmed. A deployment is a robot doing paid work, shift after shift, with an owner who counts the cost. The gap is large. Here is what we could find, with who said it.


Agility Robotics' Digit at GXO


Agility and the logistics company GXO say Digit has moved more than 100,000 totes at GXO's facility in Flowery Branch, Georgia, and that Agility has logged more than 65,000 hours of real-world operation. In October 2026 the two announced plans to extend Digit to more GXO sites, with deployments expected to start in early 2027 (reported by The Robot Report). These are the companies' figures.


The most useful sentence in that coverage comes from GXO itself. On its August 2026 earnings call, GXO's CEO Patrick Kelleher said: "We have not achieved ROI on humanoids yet," and that GXO is "a couple of years away" from that. That is a customer with a robot in a building, saying that it has not paid for itself.


Figure at BMW


Figure AI published a report in November 2025 on an 11-month deployment of its Figure 02 robots at BMW's Spartanburg plant. Figure reports 10-hour shifts, Monday to Friday, with 90,000+ parts loaded, 1,250+ hours of runtime and a contribution to production of 30,000+ BMW X3 vehicles. It also said minimal hardware failures occurred, named the forearm as its top failure point, and said it was retiring Figure 02 in favour of Figure 03. This is Figure's own report. We did not find a BMW statement confirming those figures or a follow-on contract. Earlier, in 2024, BMW had described a test at the plant and said there were no Figure robots working there at that time.


Tesla's Optimus


Reporting after Tesla's first-quarter 2026 earnings call said Optimus production at Fremont was set to begin in late summer 2026, that Elon Musk expected early output to be "quite slow", and that early skills would be "simple skills in the factory." The same coverage noted that in January 2026 Musk had acknowledged no Optimus robots were doing useful work in Tesla's factories. We found no confirmed outside sales.


What the pattern says


A few things stand out across these cases:


  • The documented work is narrow and structured: moving totes, loading sheet-metal parts, in controlled industrial buildings.
  • The numbers come mostly from the makers, and the customers are cautious.
  • Reliability and cost are the open questions, not whether a robot can do a motion once.

We are not saying these are not real. They are early.


What is hype


  • "A humanoid in every business next year." Nothing we found supports a timeline like that for small companies.
  • General-purpose labour. The deployments above are single, well-defined tasks. A robot that can do whatever a tradesperson does on a varied job site is a research goal, not a product.
  • Unit counts and prices from blogs. Several figures circulating online do not match company filings or reporting. Check the primary source.
  • "Replaces your team." Even in the best documented case, the customer says the return on investment has not arrived.

What it could mean for a physical-work business


Trades and construction. The sites change every day, the work is varied and safety is critical. A humanoid doing your install is a long way off. What helps sooner is the software around the work: quotes and estimates from photos and notes, scheduling, parts ordering, safety paperwork.


Distribution and warehousing. This is where the evidence is. Repetitive tote and pallet movement in structured buildings is where pilots are happening, often on a robots-as-a-service basis. Today's more established options are conveyors, autonomous mobile robots and fixed automation. A smaller operation can ask whether those would pay back before looking at humanoids.


Field service. Technicians go to many different homes and buildings. The nearest-term gains are in dispatch, routing, job notes, invoicing and follow-ups, which are digital.


What this means for your business


1. Do not wait for robots to fix a digital problem. Most of the delay in a trades or distribution company is in quotes, scheduling, paperwork and follow-up. That is software work you can do now.

2. If you evaluate a robot, ask for a customer who runs one. How many hours, on what task, with what failures, and what return?

3. Ask who is liable and what the safety standard is. Physical systems raise safety and insurance questions that a chatbot does not.

4. Price the whole thing: the robot, integration, maintenance, downtime, and the people who supervise it.

5. Get your data in order first. Any physical system, like any digital one, depends on clean records of jobs, parts, locations and schedules.


How this shows up at Humonix


We do not sell or install robots. Our humanoid page is a waitlist and a readiness assessment, and the physical capabilities on it are labelled Coming, not available. We say that we would help a business choose and coordinate an integrator, not that we are one.


What runs on the digital side is the work around the physical work, with each feature on a module page labelled Live, Early access or Coming: jobs and dispatch, estimating, project coordination and safety documents, kept in Know, with Hue drafting and a person approving anything that goes out. See The Agentic Enterprise.


Sources



This post is general information, not technical, safety or investment advice.