Warehouse jobs are being turned into robot training data at roughly $300 a glove, and the going rate for the person inside the glove is a paycheck they were already owed. Genesis AI raised $500 million this week to keep the recordings coming — money paid to the company, not to the hands that do the teaching.
The round closed at a $3.5 billion valuation and was announced on November 17, with Khosla Ventures, Premji Invest and Eclipse leading a list of backers who are not buying a warehouse robot. They are buying a data method. TechCrunch’s Tim Fernholz got the demonstration at the company’s Silicon Valley facility and watched two arms pack, tape and label a box — then watched the same setup drop a carton and fight a strip of printer tape on the second pass.
The robot being funded is called Eno: wheeled, two-armed, built for factories, warehouses and laboratories. Its hands carry 20 actively driven degrees of freedom and are back-drivable, meaning they give way under outside force instead of fighting it, per Tech Times’ review of the disclosed specifications. On paper that is the hardest part of the humanoid problem solved. In practice, Tech Times notes that the hands survive only weeks of regular work, that 30 to 40 Eno units exist, and that the demonstration TechCrunch attended did not use Eno at all — it used a pair of off-the-shelf arms. The $500 million is a bet on a data pipeline, priced as if the hardware question were already settled.
Zhou Xian — Dexterity Cannot Be Bought, So It Is Rented by the Hour
Genesis chief executive Zhou Xian told TechCrunch his company had “cracked dexterous manipulation,” and the way it got there is the story. Instead of paying operators to drive robots through tasks — teleoperation, where every hour of training data costs a paid human hour plus a robot — Genesis hands workers a glove that maps one-to-one onto a human hand and then one-to-one again onto Eno’s hand. More than 100 workers wear them, with headset cameras, while doing the ordinary tasks of their ordinary jobs. The gloves carry tactile electronic skin, so the recording includes touch, not just joint angles. The output feeds a model the company calls GENE-26.5.
The economics are the pitch, and the pitch is frank. A pair of gloves costs about $300. A teleoperated robot costs an operator, a robot, and a building. Genesis also claims the approach cuts hardware cost a hundredfold and improves data efficiency fivefold, and Tech Times could find no measurement behind either number. Treat those as brochures until someone publishes the arithmetic.
What is not a brochure is who does the teaching. The people generating the training set are the same people the training set is designed to retire, and they are being compensated for their shift, not for the asset they are creating.
Warehouse Jobs — The Glove Is Paid For, the Motion Is Kept Forever
Read the arrangement plainly. A worker is handed a tool that captures the exact choreography of their hands — the wrist rotation that seats a box, the fingertip pressure that keeps a label from tearing — and that capture becomes a permanent asset of a company valued at $3.5 billion. The worker’s compensation does not change. There is no royalty, no equity, no clause that says the motion stays theirs. They are, in the most literal sense available, the source material.
Construction crews do not hand over the blueprints of their technique for the price of a nail gun. This industry does it with a glove and calls the glove a gift.
The Guardian spent a day inside an Amazon fulfilment centre outside Luton this week for a piece on exactly this transition, and the description of the floor is the description of the market. Software decides where an item is stowed, which carton it ships in, how much tape seals it. The people supply the hands. Amazon employs more than 1.5 million people worldwide, and the article’s central point is that dexterity, not intelligence, is the last thing keeping that number that size.
Rodney Brooks — Touch Is the Wall, and a Glove Is a Way Around It
The roboticist Rodney Brooks has argued for years that dexterity and touch sensing remain unsolved, and that the humanoid bubble would burst on the gap between demonstration and reliability. Genesis’s glove is a direct answer to the touch half of that complaint: you cannot buy tactile data from the world, so you record it off a human hand and copy it. That is clever engineering. It is also an admission that the path to a general-purpose robot runs straight through a human being with a job.
Alistair Hunter, a hand and wrist surgeon at University College London Hospitals, put the size of the target to the Guardian: 27 bones, plus nerves and muscles, plus a brain adjusting millisecond by millisecond to what the hand is feeling. Elon Musk’s version is blunter — to build a general humanoid you must solve the hands problem. Solving it by recording 27 bones’ worth of behavior from warehouse workers is not a shortcut around the problem. It is the problem, outsourced.
Tesla’s Line Workers — They Knew What the Suit Was For
None of this is speculative, because the same method already ran at Tesla. Ars Technica, working from The Information’s reporting, describes factory workers in Texas and California wearing motion-capture suits on shift to record how they worked, and workers complaining specifically because they understood the robots were being built to replace them. Tesla’s response was to stop using line workers and move data collection to dedicated teams and training hubs.
The obvious question about the Genesis arrangement is whether its 100-plus glove wearers are permanent Genesis employees, contractors, or somebody else’s workforce on loan. No source reviewed here answers that. What is answered is that Tesla’s version produced the objection immediately, and that Tesla’s fix was to stop asking.
Also worth holding onto: Tesla assembles each Optimus hand and forearm by hand from more than 100 small parts, its touch sensors were unreliable enough that the company built a replaceable glove-like sensor layer, and Genesis’s own hands reportedly last weeks. The robots are being taught by humans and then repaired by humans. At some point the payroll you are trying to cut is the payroll keeping the product alive.
A 60-Year-Old Picker — $110,000 of Pay and Nothing Behind It
Here is where the glove lands. A worker who has spent decades picking, packing and loading hears about a $300 sensor glove differently than an investor does. 24/7 Wall St ran the arithmetic on a 60-year-old earning $55,000 a year: if the job goes now, Social Security cannot start before 62, which leaves two years and $110,000 of wages with no replacement lined up. The same analysis prices the trap at the other end — for anyone born in 1960 or later, full retirement age is 67, and claiming at 62 permanently cuts the check by about 30%. A $2,800 monthly benefit becomes $1,960, a gap of $10,080 a year that compounds across a retirement.
That is the trade being financed with $500 million. The upside is booked as a valuation. The downside is a 60-year-old learning that the last skill he has is the one the industry most needs to copy, and that copying it is the fastest way to make him unnecessary. The math says $201,600 of lost benefits across twenty years. The math does not say who is supposed to feel good about that.
Enough to Reach the Rung — Every Task Recorded Is a Task Retired
Genesis says it will begin customer deployments at the end of this year and mass-produce in 2027, with LG CNS among the names testing Eno. That gives the glove wearers roughly one production cycle of usefulness, which is precisely how long the arrangement is designed to last. Every task they record is a task in the retirement queue, and the queue is being filled by the people at the front of it.
There is a version of this that ends well, where the same dexterity that empties a picking aisle also does the lifting that put a 60-year-old’s back out in the first place. That version requires someone to decide that the humans who generated the data get a claim on what it earns. Nobody in this round signed anything that says so. The glove fits everyone the same, and only one side of the transaction sends an invoice.
Sources: TechCrunch, Tech Times, The Guardian, Ars Technica, 24/7 Wall St.