← Articles
Read in another language
Technology

Humanoid Robots Enter the Workforce: Hype vs. Reality

Jayden

Analyzes global supply chains, industrial policy, and technology issues.

Published

Key points

  • The best-documented case is Figure 02 at BMW's Spartanburg plant: over roughly 10 months in 2025 it supported production of more than 30,000 X3 vehicles, moved more than 90,000 parts, walked about 1.2 million steps and logged about 1,250 operating hours on 10-hour shifts, five days a week.
  • Agility Robotics' Digit is in commercial service with named customers — Schaeffler, GXO, Toyota Motor Manufacturing Canada and Mercado Libre — with more than 65,000 cumulative hours across nine customer facilities.
  • Tesla has never disclosed an audited deployment or production figure for Optimus, and on the Q4 2025 earnings call on January 28, 2026, Musk said it is "not in usage in our factories in a material way."
  • Reported 2025 Optimus output was a few hundred units — less than 10% of the 5,000-unit target — against earlier talk of building roughly 10,000 units that year.
  • The verified work is real but narrow: sheet-metal picking and tote moving, at limited unit counts per site, which is why "humanoids have entered the workplace" is both true and overstated.

In 2026, human-shaped robots finally walked out of the demo video. At BMW's Spartanburg plant in South Carolina, a humanoid robot (a robot built to resemble the human body) from Figure AI carried parts on a real vehicle line and logged operating hours, and BMW's Leipzig plant in Germany followed. Around the same time, Tesla's Elon Musk kept saying his company's humanoid, Optimus, would "soon fill the factories." Yet on one side there is a documented record of hours worked, and on the other, a promise that has not yet been verified.

Why is the whole world watching this now? Whether humanoids are truly entering the workplace carries high stakes for automation, employment, and investment. But the sentence "humanoids are working" can mean a target a company has announced, or a fact that has been independently confirmed. This article does not blend the two; it separates, as of 2026, the verified deployments from the claims that remain announcements, to gauge where humanoid commercialization actually stands.

A word on method before the cases. Throughout this piece we keep three kinds of statements apart: what has been verified (operating logs, named customers, counted hours that a third party can check), what a company has announced (its own descriptions of its products, order books, and roadmaps, which may well be true but have not been independently confirmed), and what has been forecast (analyst scenarios about how many robots will be working in some future year). All three appear in humanoid coverage, usually in the same sentence. Sorting them is not pedantry — it is the only way to tell a shift that is already happening from one that has only been promised.

Table of Contents

  1. What is "actually" deployed in 2026
  2. The verified floor — Figure and BMW
  3. Another verified case — Agility's Digit
  4. Between announcement and verification — Tesla Optimus
  5. How to read the numbers — why verification matters
  6. What to watch

What is "actually" deployed in 2026

From stage demo to shift work

For a few years, humanoid robots have drawn attention through flashy stage videos. Clips of them climbing stairs, dancing, and picking up eggs circulated on social media. But a stage demo and daily factory labor are different stories. What makes 2026 notable is that a handful of humanoids began to be described in the language of "work" rather than "demonstration" — labor measured not in days of events but in shifts, operating hours, and the number of parts moved.

The difference between those two vocabularies is larger than it sounds. A demonstration is staged once, under conditions its makers chose, and its success is judged by whether the robot completed the trick. Work is repeated, under conditions the factory chose, and its success is judged by whether the line kept moving. That is why the arrival of shift language matters: shifts, operating hours, and part counts are the units a plant already uses to describe its human workers, and applying them to a robot implies the robot was inside the same system of measurement rather than beside it.

The question that separates the two

The key point is that the word "deployed" is slippery. At one company it means a documented operating log; at another it means a roadmap that will supposedly get there soon. So as you read each case below, one question exposes the reality: can you confirm, in numbers, who did what, where, for how long? Let us separate deployments that can answer this from announcements that cannot yet.

That single question unpacks into four smaller ones, and they are worth holding in mind through every case that follows. Is there a named operator or customer, rather than an unnamed "leading manufacturer"? Is there a specific task in a specific environment, rather than a category like "logistics"? Is there a duration — hours, shifts, months — rather than an adjective like "extensively"? And is there a unit count, so that "deployed" means something other than "at least one"? A claim that answers all four can be checked by someone outside the company. A claim that answers none of them is a press release, however impressive the video attached to it.

The verified floor — Figure and BMW

The measured log at Spartanburg

The best-documented case is the collaboration between the U.S. startup Figure AI and BMW. In 2025, the Figure 02 robot deployed at BMW's Spartanburg plant became the first humanoid to operate at a BMW facility anywhere in the world. By BMW's own count, over roughly ten months the robot supported the production of more than 30,000 BMW X3 vehicles, moved more than 90,000 components, took about 1.2 million steps, and logged around 1,250 operating hours. It worked ten-hour shifts, five days a week, and its task was to precisely pick up sheet-metal parts used in the welding process and place them where they belonged [source: BMW Group, 2025].

These numbers matter because they are measured, not impressionistic. Not "the robot helped," but "it moved this many parts, over this many hours, in this daily pattern," recorded as a log.

Notice in particular that the shift pattern is reported alongside the totals. A headline number on its own — "90,000 components" — could describe almost any span of time. Paired with ten-hour shifts, five days a week, and a roughly ten-month window, it describes a routine: the robot was scheduled, not demonstrated. This is also why the Spartanburg figures survive scrutiny better than most humanoid claims. They come from the plant operator rather than the robot maker, they are expressed in the plant's own accounting units, and they name the exact task — picking and positioning sheet-metal parts for welding — so that anyone can judge how demanding the work actually was.

What Figure 03 changes — and what it does not

In 2026, the follow-up came: the next-generation Figure 03 began to be deployed at the same Spartanburg plant. Its task this time is to pick up logistics parts and sort them in order into a sequencing trolley, and improvements such as soft outer components for safety, wireless charging, and hands with tactile sensors were introduced [source: BMW Group / Figure AI, 2026]. Here, though, a distinction is needed. The number of Figure 03 units deployed has not been officially disclosed, and this is an early-stage deployment rather than full production. What is verified stops at "a new robot has started a new task"; the scale cannot yet be stated in numbers.

The sequence behind that follow-up is worth spelling out, because it shows how slowly this moves even in the best-documented case. BMW describes the Figure 03 deployment, announced in June 2026, as coming after roughly a year of body-shop validation with the previous generation — the robot did not jump from pilot to a new assignment in a matter of weeks [source: BMW Group / Figure AI, 2026]. The feature list also includes voice-to-voice conversation alongside the soft outer components, wireless charging, and tactile-sensing hands [source: BMW Group / Figure AI, 2026]. Those are capabilities the makers describe, though, not results a plant has logged, which places them one tier below the Spartanburg operating record. A year of validation to move one generation of robot onto one new task is the realistic clock speed of this technology on a working line.

Germany's first humanoid deployment

BMW's experiment does not stay in the United States. In 2026, BMW deployed a humanoid robot named AEON as a pilot at its Leipzig plant in Germany, creating the first humanoid deployment in the country. After an initial test in December 2025, the robot entered its pilot phase in the summer of 2026, handling high-voltage battery assembly and component manufacturing. BMW calls this trend "Physical AI" (artificial intelligence that operates in the physical world) and established a dedicated unit in Munich, the "Center of Competence for Physical AI in Production" [source: BMW Group, 2026]. In short, the BMW-Figure case is the best-verified humanoid labor on an industrial floor so far — but it is worth remembering that most of it is still described in the language of pilots and single tasks.

Two details about Leipzig complicate the picture in useful ways. First, AEON comes from Hexagon rather than Figure, and it is a wheeled mobile platform that can take attachable grippers and scanners rather than a walking machine [source: BMW Group, 2026] — a reminder that "humanoid" in industrial practice covers a range of body plans, and that legs are a design choice rather than the definition. Second, the deployment advanced in careful stages: initial tests in December 2025, further testing from April 2026, and only then the pilot phase in the summer of 2026 [source: BMW Group, 2026]. Read together, they suggest that what BMW is validating is not a single robot but a way of introducing robots — which is also why the Munich competence center matters more than any one machine on the floor.

Another verified case — Agility's Digit

Named customers, counted hours

BMW-Figure is not the only verified case. Digit, the two-legged robot made by Oregon-based Agility Robotics, is running under commercial contracts in logistics and manufacturing settings. According to Agility's official announcement in June 2026, Digit is used by customers such as Schaeffler, GXO, Toyota Motor Manufacturing Canada, and Mercado Libre, and has accumulated more than 65,000 hours of operation across nine customer facilities in real production environments [source: Agility Robotics, 2026]. The phrase to note here is "real production environments." Because these are commercial deployments with named customers and stated operating hours, this too is closer to verification than to announcement.

Secondary accounts of individual customer sites fill in what those hours look like in practice. GXO is described as having signed the industry's first formal commercial humanoid robotics-as-a-service contract in 2024, and Digit is reported to have moved more than 100,000 totes at GXO's Flowery Branch facility in Georgia through November 2025. Toyota Motor Manufacturing Canada is reported to have deployed seven Digit units at its Woodstock, Ontario plant, which builds the RAV4, in February 2026 under the same subscription model. Amazon tested Digit at an R&D facility in 2023 on tote recycling and is an investor through its Industrial Innovation Fund. These are secondary accounts rather than audited disclosures, so they belong one tier below the headline operating hours — but they are specific about site, date, and unit count, which is more than most humanoid claims manage.

What that specificity buys is the ability to be wrong in public. If a named customer at a named facility were not running the robots described, the customer could say so. That accountability, more than any single number, is what separates this tier of evidence from a product page. It is also why the unit counts deserve attention: seven robots at one plant is a real deployment and a small one at the same time, and both halves of that sentence are true. The verified picture at Agility is of a technology that has genuinely entered commercial service, at a scale still measured in single-digit and double-digit units per site.

The announcement layer at the same company

Agility used this track record to go public as well. Through a merger with a special-purpose acquisition company (SPAC), it is pursuing a listing at a $2.5 billion pre-money valuation under the ticker AGLT, and it introduced its latest model, Digit v5, as "the world's first AI-enabled cooperatively safe humanoid" (designed to work safely alongside people). The company said it has secured more than $300 million in multi-year orders for Digit v5 and a pipeline of more than 30 customers [source: Agility Robotics, 2026]. That said, these order and pipeline figures carry the character of a company announcement, and how much converts into actual revenue has not yet been independently verified. Moreover, Digit's verified work is largely concentrated on repetitive logistics, such as moving empty boxes (totes), which is a long way from the narrative that "humanoids will replace human labor across the board."

The transaction itself sits in the same announcement layer. The merger partner is Churchill Capital Corp XI, and the deal was expected to close during 2026 [source: Agility Robotics, 2026] — an expectation rather than a completed event at the time of the announcement. This is the cleanest illustration of why tiers matter within a single company: the 65,000 hours and the named customer sites are the record, while the valuation, the order book, the pipeline, and the closing timetable are all statements about a future that has not happened yet. Both sets of numbers appeared in the same press release. Only one of them describes work that has already been done.

Between announcement and verification — Tesla Optimus

What has never been disclosed

Now consider the opposite end of the spectrum. The thing that most inflated public expectations for humanoids is Tesla's Optimus — but from the standpoint of verification, the story changes. Tesla has never disclosed an audited official figure for how many Optimus units have been deployed or produced, and no commercial sales have taken place.

The absence is worth dwelling on, because absences are easy to read past. Tesla is a public company that reports production and delivery figures for its vehicles every quarter in considerable detail. The same disclosure machinery that produces those numbers has never produced an Optimus unit count. That does not prove nothing is being built; it does mean that anyone claiming a specific number of working Optimus robots is relying on something other than the company's audited reporting. When a firm that counts everything declines to count one thing, the silence is itself information.

The company's own correction

More decisive is the company's own admission. On January 28, 2026, in the Q4 2025 earnings call, Musk said: "We are still very much at the early stages of Optimus. It's still in the R&D phase. We have had Optimus do some basic tasks in the factory. But as we iterate on new versions of Optimus, we deprecate the old versions. It's not in usage in our factories in a material way. It's more so that the robot can learn" [source: Tesla Q4 2025 earnings call, 2026]. In other words, the CEO himself stated that Optimus is not yet doing productive work.

A self-correction of this kind is unusually strong evidence, and it is worth being explicit about why. Companies have every incentive to describe their own products generously; the pressure on an earnings call runs toward optimism, not caution. A statement that cuts against that pressure — made by the chief executive, on the record, in the venue where investors are listening — is the rare claim that does not need independent confirmation to be credible, because the party making it had nothing to gain from it. It also settles the question at issue here without appeal to outside analysis: the argument that Optimus is not yet doing verified factory work is not a skeptic's inference. It is the company's own account.

A roadmap that keeps moving

This remark stands out precisely because it contrasts with earlier pronouncements. At the June 2024 shareholder meeting, Musk said he expected "a thousand to a couple thousand robots working in its factories" by 2025, and in early 2025 he paired a plan to build roughly 10,000 units that year with the claim that he was "confident" several thousand robots would be doing useful things by year's end. Yet according to reporting, actual 2025 production came in at a few hundred units — less than 10% of the 5,000-unit target [source: Electrek, 2026]. The roadmap remains aggressive. Tesla's Q1 2026 shareholder update said Gen 3 hand production began on January 21, 2026, and that full-body volume production is targeted for late July or August 2026 at the Fremont plant. That line is designed with an eye toward one million units a year, though Musk himself added that initial output would be "quite slow" and that this year's production rate is "literally impossible to predict" [source: Tesla Q1 2026 Shareholder Update, 2026]. The gap between the size of the target and the verified track record is exactly what makes Optimus a case of "announcement."

Two further details from that update show how much the company is committing, and how much remains a target. The Fremont volume line is to take over space from the Model S and Model X lines, and deliveries to outside customers are aimed at the end of 2026 [source: Tesla Q1 2026 Shareholder Update, 2026]. Converting existing vehicle capacity is a substantial internal commitment — it is not the behavior of a company that has quietly shelved the program. But every element here belongs to the projection tier: a target start month, a target customer-delivery window, and a design capacity for a line that has not yet run at volume. Held against the CEO's own "impossible to predict," the honest summary is that Tesla's Optimus effort is large, funded, and unverified all at once, and only an auditable unit count will move it out of that category.

How to read the numbers — why verification matters

Three tiers of claim

Placing these cases side by side reveals three tiers for reading humanoid news. First, verified measurements — third-party-checkable numbers like BMW-Figure's operating logs and Agility's customers and hours. Second, company announcements and marketing — in-house phrasing like "next-generation," "world's first," and order pipelines, which may be true but has not yet been independently verified. Third, forecasts and estimates — scenarios about the future, like analysts predicting humanoids will "soon replace millions of workers." Read all three with equal weight and you lose the reality.

The practical value of the tiers is that they tell you what a claim is for. A verified measurement can settle an argument about what is happening now. An announcement tells you what a company intends and is willing to be judged on later, which is genuinely informative — a firm that publishes an order book has staked something — but it cannot settle a present-tense question. A forecast tells you what someone believes about a year that has not arrived, and its accuracy is unknowable until it does. The error that produces most of the confusion in this field is not believing any one of the three; it is quoting a tier-two or tier-three number in a sentence written in the present tense.

True and overstated at the same time

So the proposition "humanoids have entered the workplace" is both true and overstated. It is true because documented operation genuinely exists, as in the BMW and Agility cases; it is overstated because that labor still sits within narrow, repetitive tasks such as sheet-metal picking or tote moving, and the unit counts are limited. At the same time, the opposite skepticism — that "it's all just PR" — is also inaccurate. In at least two places, it is shifts, not press releases, that are underway. Fair reporting begins by acknowledging both. The real question is not "are humanoids working?" but "across how wide a range of tasks, at what scale, and how sustainably?" — and most of that answer is still open.

Holding both halves at once is harder than it sounds, because each half has a constituency. The optimistic reading has companies and investors behind it and a real operating record to point at; the dismissive reading has the long history of overpromised robotics behind it and a real record of missed targets to point at. Neither is wrong about its own evidence. What each gets wrong is the scope: the boosters generalize a handful of documented tasks into a labor-market transformation, and the skeptics generalize one company's missed roadmap into an industry-wide illusion. The defensible position is narrower and duller than either — a small number of robots are doing a small number of tasks, and they are doing them for real.

What to watch

What 2026 established

To sum up, 2026 is the year humanoids moved beyond demonstration into verified labor. The verified facts are these: Figure 02 supported the production of more than 30,000 X3s at BMW Spartanburg and left a documented operating log, followed by the next-generation Figure 03 and the Leipzig pilot in Germany. Agility's Digit has run commercially for more than 65,000 hours at named customers. Tesla's Optimus, by contrast, has no independently verified deployment record despite an aggressive roadmap, and its CEO conceded it is not used in the factory in a material way.

Three things that would settle it

Three things are worth watching from here. First, whether Tesla's Fremont line produces an auditable unit count that turns its roadmap into verification. Second, whether pilots like Figure 03 and BMW Leipzig move beyond single-task demonstrations into sustained production across multiple shifts. Third, whether the economics of Digit-style robot subscription (RaaS, or Robotics-as-a-Service, renting a robot like a service) scale beyond repetitive logistics into wider tasks. Humanoids have clearly crossed the factory threshold. But whether that is a handful of pilot cases or a shift that becomes the industry standard will be answered not by announcements, but by the verified numbers of the next few years.

It is worth naming in advance what an answer would look like, so that the next round of news can be judged rather than absorbed. A verification-grade disclosure would identify the operator, the site, and the task; report a duration in hours or shifts; state how many units were involved; and come from the party that runs the floor rather than the party that sells the robot. Every claim in this article that survived scrutiny had most of those elements, and every claim that did not survive was missing them. That is the whole test, and it does not require any special expertise to apply — only the discipline to ask for the numbers before accepting the sentence.

Charts

Operating hours on the record

Operating hours on the recordFigure 02 at BMW Spartanburg (2025) 1,250hours, Agility Digit across 9 customer facilities 65,000hours1,250hoursFigure 02 at BMW Spartanburg (2025)65,000hoursAgility Digit across 9 customer facilities
Two different kinds of count: about 1,250 hours is one robot's log at a single BMW plant over roughly 10 months (BMW Group), while more than 65,000 hours is Agility's cumulative total across nine customer facilities (Agility Robotics). Both are disclosed operating records rather than projections.

Items moved in documented deployments

Items moved in documented deploymentsParts moved by Figure 02 at BMW Spartanburg (verified) 90,000items, Totes moved by Digit at GXO Flowery Branch (secondary account) 100,000items90,000itemsParts moved by Figure 02 at BMW Spartanburg (verified)100,000itemsTotes moved by Digit at GXO Flowery Branch (secondary account)
Both figures are floors — the sources say "more than" — and the items differ: sheet-metal and other parts at BMW versus warehouse totes at GXO through November 2025. The BMW number comes from the plant operator; the GXO number is a secondary account rather than an audited disclosure.

Tesla Optimus: announced targets for 2025

Tesla Optimus: announced targets for 2025Expected in factories, said June 2024 (high end) 2,000units, 2025 production target 5,000units, Units to build in 2025, said January 2025 10,000units2,000unitsExpected in factories, said June 2024 (high end)5,000units2025 production target10,000unitsUnits to build in 2025, said January 2025
All three are announcements for the same year, made at different times. Reported actual output was a few hundred units — less than 10% of the 5,000-unit target — and Tesla has never published an audited figure, so no measured bar can be plotted here.Tesla and Elon Musk statements, as reported by Electrek (opens in a new tab)

Timeline

  1. At Tesla's shareholder meeting, Musk says he expects "a thousand to a couple thousand robots working in its factories" by 2025.

    Elon Musk, shareholder meeting, as reported by Electrek (opens in a new tab)
  2. Figure 02 works about 10 months at BMW Spartanburg — the first humanoid at a BMW facility — moving more than 90,000 parts across about 1,250 operating hours while supporting production of more than 30,000 X3 vehicles.

    BMW Group (opens in a new tab)
  3. By this point, secondary accounts put Digit's total at more than 100,000 totes moved at GXO's Flowery Branch, Georgia facility, under an industry-first commercial humanoid RaaS contract signed in 2024.

  4. BMW begins initial tests of the AEON humanoid at its Leipzig plant — the first humanoid deployment in Germany.

    BMW Group (opens in a new tab)
  5. Tesla begins production of the Gen 3 Optimus hand.

    Tesla — Q1 2026 Shareholder Update (opens in a new tab)
  6. On the Q4 2025 earnings call, Musk says Optimus is still in the R&D phase and "not in usage in our factories in a material way."

    Tesla Q4 2025 earnings call, via Electrek (opens in a new tab)
  7. Secondary accounts report that Toyota Motor Manufacturing Canada deploys 7 Digit units at its Woodstock, Ontario plant, which builds the RAV4, under a robots-as-a-service subscription.

  8. Further tests of AEON run at BMW Leipzig from April onward, ahead of the pilot stage.

    BMW Group (opens in a new tab)
  9. BMW announces that the next-generation Figure 03 has started work at Spartanburg on logistics sequencing, after roughly a year of body-shop validation with the previous generation. The number of units is not disclosed.

    BMW Group / Figure AI (opens in a new tab)
  10. Agility Robotics reports more than 65,000 operating hours across nine customer facilities and announces a SPAC merger with Churchill Capital Corp XI at a $2.5 billion pre-money valuation, ticker AGLT, with closing expected within 2026.

    Agility Robotics (opens in a new tab)
  11. Targeted, not achieved: Tesla aims to start full-body Optimus volume production at Fremont, taking over space from the Model S and Model X lines, with outside-customer deliveries aimed at the end of 2026.

    Tesla — Q1 2026 Shareholder Update (opens in a new tab)

Analysis

"Deployed" is not one word

At one company it means a documented operating log; at another it means a roadmap toward one. The whole difference between the cases in this story sits inside that single verb, which is why the same headline can be accurate about BMW and misleading about Tesla.

Who counts matters as much as what is counted

The Spartanburg figures came from the plant operator rather than the robot maker, and they are expressed in the plant's own accounting units — shifts, operating hours, parts moved. That is why they survive scrutiny better than most humanoid claims.

A self-downgrade is unusually strong evidence

Companies face pressure to describe their own products generously, especially on an earnings call. A statement that cuts against that pressure — from the CEO, on the record — is one of the few claims that is credible without independent confirmation, because the party making it gains nothing by it.

Legs are a design choice, not a definition

AEON at Leipzig comes from the Hexagon lineage and is a wheeled mobile platform with attachable grippers and scanners rather than a walking machine. "Humanoid" on a factory floor already covers several body plans.

Comparison

Three tiers of claim — keeping them apart is the article's central discipline.
TierExamples in this storyWhat it is
VerifiedFigure 02 at BMW Spartanburg: ~1,250 hours, 90,000+ parts, 30,000+ X3 over ~10 months; Agility Digit: 65,000+ hours across 9 customer facilitiesOperating records disclosed by the plant operator or counted across named customers
AnnouncedFigure 03's features and its June 2026 rollout; AEON's Leipzig pilot; Digit v5 as "the world's first cooperatively safe humanoid," $300M+ in multi-year orders, a 30-plus customer pipeline, a $2.5B pre-money SPAC valuationA company's own statements about its products, orders and plans
ProjectedVolume production targeted for late July or August 2026 at Fremont; one million units a year of line design capacity; deliveries to outside customers aimed at end-2026; analyst forecasts of humanoids replacing millions of workersStatements about a year that has not arrived yet
Case by case: what each deployment has actually established, and what is still an announcement.
DeploymentVerifiedStill announced or projected
Figure 02 — BMW Spartanburg (2025)~10 months, 30,000+ X3 supported, 90,000+ parts, ~1.2M steps, ~1,250 hours, 10-hour shifts five days a week, picking and positioning sheet metal for weldingWhether the pattern extends beyond a single task
Figure 03 — BMW Spartanburg (2026)A next-generation robot began logistics sequencing after roughly a year of body-shop validation with the previous generationUnit count not disclosed; an early-stage deployment, not full production; soft outer shell, wireless charging, voice-to-voice conversation and tactile-sensing hands are maker descriptions
AEON — BMW Leipzig (2026)Germany's first humanoid deployment, staged from initial tests in December 2025 through further tests from April 2026; assigned to high-voltage battery assembly and component manufacturing; a wheeled platform with attachable grippers and scannersStill described as a pilot entering its stage in summer 2026; no operating log published
Digit — Agility Robotics65,000+ hours across nine facilities of named customers, under commercial contracts$300M+ multi-year orders, 30-plus customer pipeline, $2.5B pre-money valuation, SPAC merger with Churchill Capital Corp XI expected to close within 2026
Optimus — TeslaNothing: no audited deployment or production figure, no commercial sales, and a CEO statement that it is "not in usage in our factories in a material way"Full-body volume production targeted for late July or August 2026 at Fremont, taking over space from the Model S and Model X lines; one million units a year of design capacity; outside-customer deliveries aimed at end-2026

Process

  1. A named operator or customer

    Not "a leading manufacturer" but a company that could publicly contradict the claim if it were false.

  2. A specific task at a specific site

    Not a category like "logistics" but work anyone can picture, such as picking sheet metal for welding.

  3. A duration in hours, shifts or months

    Not an adjective like "extensively" — the unit a plant already uses for human workers.

  4. A unit count

    So that "deployment" means something other than "at least one."

Sources

  1. BMW Group — "BMW Group advances the use of Physical AI in production with Figure 03 project in Spartanburg" (2026).View source (opens in a new tab)
  2. BMW Group — "BMW Group to deploy humanoid robots in production in Germany for the first time" (Leipzig pilot and Spartanburg results, 2026).View source (opens in a new tab)
  3. BMW Group — "Humanoid Robots for BMW Group Plant Spartanburg" (background, 2024).View source (opens in a new tab)
  4. Agility Robotics — "Agility Robotics to Go Public Through Merger with Churchill Capital Corp XI" (2026-06-24).View source (opens in a new tab)
  5. Tesla — Q1 2026 Shareholder Update (Optimus production roadmap, 2026).View source (opens in a new tab)
  6. Tesla Q4 2025 earnings call / Electrek — "Musk admits no Optimus robots are doing 'useful work' at Tesla" (2026-01-28).View source (opens in a new tab)
  7. Technology.org — "Humanoid Robots in 2026: What Is Actually Deployed" (2026-07-18).View source (opens in a new tab)

Tags

  • #humanoid-robots
  • #figure-ai
  • #tesla-optimus
  • #bmw
  • #agility-robotics
  • #factory-automation