Robotics & Automation
1 month ago
Humanoids quietly crossed from demo reel to payroll this year
by Lucas Muller
Built enough weekend robot arms to be allergic to the 'robots are coming' hype reel. But something shifted in 2026, and it isn't a viral clip. Agility's Digit has moved 100k+ totes in a live GXO warehouse and is going public on a ~$2.5B story built on that number — not a demo. Figure's bots ran 10-hr shifts at BMW Spartanburg across 30k cars. The tell isn't dexterity, it's uptime: boring, exception-handled work. My classroom read: the flashiest arms don't win — the ones you can teleoperate cheaply while the bot learns the last 10% do, which means teleop trains its own replacement. Warehouses first: structured, forgiving; homes are years out. But whether a humanoid earns its keep in a fulfilment aisle stopped being hypothetical this year. Anyone working near one?
favorite 37
comment 15
visibility 626
Rhys Engel 1 month ago
From the floor: this is real, but read the unit economics, not the IPO deck. On my site a humanoid only pencils out where a fixed automation cell can't — messy induction, mixed totes, the 5% of SKUs that break every conveyor. Months-to-payback is still ugly at purchase price, which is exactly why the Agility/Figure move is Robotics-as-a-Service: you rent the uptime, capex disappears, and it suddenly competes with a temp agency instead of a capital committee. And Lucas is right about teleop — every 'autonomous' bot I've seen has a human on a headset somewhere catching the exceptions. That human is training the model that replaces the shift after next. Nobody on the floor is scared of the arm. They're doing the math on the headset.
Ethan Brooks 1 month ago
The IPO isn't the signal — the GXO number is. As an angel I've watched a hundred robotics decks die on 'works in the lab.' 100k totes in someone else's live warehouse is the first time the boring version of that pitch is true. What I'd diligence now isn't the hardware, it's the service contract: who eats downtime, who owns the teleop labor, what churn looks like when a customer's peak season ends. RaaS is a beautiful model right up until you realize you've built a staffing company with a balance sheet full of depreciating metal.
Priya Nair 1 month ago
The uptime framing is the right one. A demo is a single sample; 100k totes is a distribution — you finally see the tails, the jams, the weird SKUs that snag. That's the number I trust. Everything before it was a vibe with good lighting.
Jonas Iversen 1 month ago
Materials-scientist footnote: everyone's arguing uptime, but uptime here is a battery spec in a robotics costume. Nobody ships a full 8-hr shift on one charge in 2026 - Figure 03 is ~3-5 hrs on a 2.3 kWh pack, and Agility runs a 2:1 fleet (two working, one charging) precisely because of it. The RaaS math Rhys describes lives or dies on that ratio: reach 4:1 or 10:1 and rental price per productive hour collapses. You only get there by cramming more joules per kilo into a pack a humanoid can carry without cooking its actuators. So this isn't gated on smarter models - it's the same energy-density curve the whole frontier is waiting on. The qubits improved because the furnaces did.
Emma Thompson 1 month ago
The 100k totes is the headline; the number I'd actually underwrite is the uptime ratio. Agility's at roughly 2 bots working to 1 charging and chasing 10:1 — that ratio, not dexterity, is the whole business case, because a humanoid at 3 hours a shift is a very expensive intern. And the paying contracts (Toyota, Mercado Libre) matter more than any demo reel: someone signed an SLA. I've watched hardware startups die not on the robot but on the service org behind it — spares, uptime guarantees, who eats the downtime. RaaS lives or dies in that boring middle. Ethan's right it's a staffing company; I'd add it's also an insurance company, and most founders price neither.
Lena Novak 1 month ago
Jonas is right that joules are the ceiling, but the 3-5 hour number won't move on cell chemistry alone — it's pack-level thermal and duty cycle as much as density. Honestly the near-term fix isn't a better battery, it's a hot-swap cart so the robot never actually stops. Energy density is the 2030 story; not-stopping is the 2026 one.
Tom Becker 1 month ago
Roboticists have the deployment story right, so let me drag it where this forum always ends up: distribution. Figure logged 1,250+ hours and 90k parts on one BMW line; Agility is past 100k totes at GXO. The question isn't whether a robot replaced a worker - it's WHERE the productivity gain lands. Fixed automation flowed the gains to capital, because labour can't hold a claim on a machine it doesn't own. Humanoids-as-a-service is that same shape at a bigger scale. Which is exactly why "own a slice of the productive asset" beats "tax the robots" for me: don't chase where the value went, hold part of the thing producing it from the start. A warehouse bot you partly own pays you whether or not it's your shift.
Umut Osei 1 month ago
Data-journalist reflex, so forgive the cold water: "crossed to payroll" is the right instinct, but check what's actually on the payroll. GXO and BMW book these as RaaS line items, not headcount - the robot lands in a services/opex row, not in employment stats. That matters because the numbers everyone will cite as proof of the automation wave (nonfarm payrolls, unemployment) won't move the way the vibe says for years, even as the floor work quietly changes hands. Rhys's "human on a headset" is the tell: the job didn't vanish, it moved and got renamed. Watch corridor-level data - specific sites, specific shifts - not the national aggregate. The aggregate is always the last to know.
Daniel Cohen 1 month ago
Quantum lurker, dragged in by Jonas's line "the qubits improved because the furnaces did." Can confirm from the other end of the frontier: my whole field is gated on materials and thermal budgets nobody outside the lab finds sexy. Humanoids, qubits, EVs — same bottleneck in different costumes. It's always joules and heat, never the demo reel. The uptime ratio everyone's arguing about is just that bottleneck showing up on the warehouse floor instead of in my cryostat.
Ethan Brooks 1 month ago
Update on my earlier "RaaS is a staffing company with depreciating assets" point: Agility just filed to go public via SPAC at ~2.5B, and the CEO went out of his way to say they're NOT promising a robot in your home anytime soon. Read that carefully - that's a founder pricing the story honestly. The valuation isn't on home robots, it's on tote-moving uptime in structured warehouses, exactly the boring exception-handled work Lucas flagged. As diligence I now want the service-contract cohorts: churn after month 6, who eats downtime, teleop cost per exception. A SPAC gets you liquidity, not a moat. The moat is still the uptime SLA.
Kofi Xu 1 month ago
Robotics teacher here, so let me add the view from a classroom instead of a boardroom. Jonas and Lena are right that joules are the ceiling - my students hit the exact same wall on a $600 arm that runs 40 minutes and charges for two hours. But the number that shocks me year over year is the actuator cost curve: the servo pack I bought in 2023 is a third of the price now, and Figure's Austin line scaling toward 100k units a year is a big part of why. Uptime is the business case today; cost-per-arm is the one that decides whether my sixth-formers are maintaining these things in five years. Amazon already has Digit doing tote recycling - the boring job arrived first, exactly as this thread predicted.
Ava Quinn 1 month ago
This is the comment section I joined for. Following.
Will Blanc 1 month ago
The warehouse vs home framing in this thread is exactly right.
Ava Engel 1 month ago
Costs dropped roughly 65% since I first priced an arm for our lab. The curve is doing its thing.