Note: This is a research note supplementing the book Unscarcity, now available for purchase. These notes expand on concepts from the main text. Start here or get the book.

Autonomy Theater: The Humans Behind “Autonomous” Robots

The demo shows a robot folding your laundry by itself. The fine print says a stranger in another time zone can grab the controls. Both are true, and the gap between them is where the whole story lives.

In 2026, Weave Robotics started taking orders for Isaac, an $8,000 home robot (or $449 a month, if you’d rather rent your butler) that folds laundry, tidies clutter, and makes your bed. The launch video is charming: the machine gathers a scattered pile of clothes, carries the hamper, and folds a shirt with the patient competence of a department-store employee ten minutes before closing.

Read Weave’s own materials, though, and a second character appears. Isaac is “autonomous by default” for its two headline routines (Weave calls them Laundry Flow and Daily Reset), but when the robot gets stuck, “teleoperation assistance may step in.” A remote human specialist can connect over cellular or Wi-Fi and take manual control of the arms, guiding the machine through whatever it couldn’t handle. To do that, the robot’s cameras capture its surroundings (including, Weave notes, “people nearby”), and the company’s workforce may remotely access that footage.

So: the robot folds your laundry. Except when a person you’ll never meet is folding it for you, through the robot, while watching your living room. That’s not a scandal. Weave discloses it. It’s something more interesting than a scandal. It’s the tell that the entire category of “autonomous machines” has been quietly running on a blend of silicon and hidden humans the whole time.

The Wizard Behind the Curtain Has a Name

The name is ghost work, coined by researchers Mary L. Gray and Siddharth Suri to describe the vast, deliberately invisible layer of human labor that props up systems marketed as automated. Their point was blunt: an awful lot of “AI” is a person in a chair somewhere, doing the part the software can’t, fast enough that you assume no one is there.

Weave is just the latest, most literal example, because now the ghost is reaching into your home and pulling on physical clothing. But the pattern is a genre at this point, and once you learn to see it, you can’t unsee it:

• Amazon “Just Walk Out.” For years, Amazon’s cashierless stores were the poster child for retail automation: grab your items, walk out, get charged automatically, no human involved. Then reporting in 2024 revealed that the “AI” leaned on more than 1,000 workers in India reviewing and labeling video of shoppers, which is part of why receipts sometimes arrived hours after you left. Amazon disputed the framing (its associates annotate footage to train the system, the company said; they don’t sit watching a live feed to ring you up), but nobody disputes that the humans were essential and that the marketing never mentioned them. Amazon has since been swapping Just Walk Out for smart carts, which at least put the sensor where you can see it.
• Cruise robotaxis. GM’s self-driving fleet looked like the future until, in late 2023, CEO Kyle Vogt confirmed the cars were remotely assisted 2 to 4 percent of the time. Reporting put harder numbers on it: a remote human stepping in every four to five miles, and roughly 1.5 staff for every car on the road. The company insisted these were “wayfinding” nudges, not joystick driving. Either way, the “driverless” car had a support team the size of a small call center, and after a pedestrian was dragged in a collision, the fleet was grounded and the robotaxi business wound down.
• Tesla Optimus. At Tesla’s October 2024 “We, Robot” event, humanoid robots mingled with the crowd, poured drinks, and bantered with guests. The robots walked on their own, but the interactions (the talking, the bartending, the charming hand gestures) were teleoperated by Tesla staff offstage. At least one Optimus, asked directly, admitted it was “assisted by a human.” The bots were real. The autonomy was theater.

Four companies, four price points, one script. Show the audience the finished trick. Keep the assistant behind the curtain.

Autonomy Is a Spectrum, Not a Switch

Here’s the mental model most people carry, and it’s wrong: a task is either automated or it isn’t. The robot can fold laundry or it can’t. The car drives itself or it doesn’t. Flip the switch, cross the line, humans out, machines in.

Real automation almost never works like a switch. It works like a dimmer, sliding from “human does everything” through a long middle band of “human supervises, machine acts, human catches the failures” before it ever reaches “machine does everything unattended.” That middle band, the human-in-the-loop stage, isn’t a bug or an embarrassing secret. It’s a predictable, engineered phase of every hard automation problem. Sometimes it lasts months. Sometimes it lasts a decade. And for some tasks, it may be the permanent equilibrium, because the last 5 percent of edge cases costs more to automate than it costs to keep a person on call forever.

This is the distinction the book keeps returning to: the difference between demonstrated automation and durable automation. A demo proves a machine can do a task once, under good lighting, with an operator ready to save it. Durable automation means the machine does the task, all day, unattended, cheaper than a human, at a reliability the world will actually tolerate. The distance between those two is exactly the distance a remote operator in Manila is currently covering with a headset and a joystick.

Which reframes the entire automation debate. The interesting question was never “are robots autonomous yet?” That yes/no flatters the demo and hides the operator. The honest questions are three, and the book is built around them:

1. Which tasks cross fully to the machine? (The dimmer goes all the way up. Nobody’s in the loop. This is the real edge of the Labor Cliff.)
2. Which tasks stay human-assisted more or less forever? (The dimmer sticks in the middle. There’s always a person on call.)
3. Who are the hidden humans in the meantime, and what do we owe them?

That last one is where the moral weight sits, and it’s the one autonomy theater is designed to make you forget.

Who’s Actually in the Loop?

The uncomfortable answer: it’s usually not the person in the launch video. The ghost workforce is disproportionately in the Global South: the reviewers in India behind Just Walk Out, the delivery-robot operators piloting sidewalk bots from Colombia, the data labelers across Kenya and the Philippines who taught the models what a stop sign and a folded shirt look like. They’re paid by the task, invisible by design, and structurally excluded from the upside of the very automation they make possible.

This is the precariat the book warns about, and it’s a nasty trap. These workers are too essential to fire but too hidden to bargain. They don’t show up in employment statistics as “robot operators,” because the whole product depends on the fiction that no operator exists. They aren’t building equity or recognized standing; they’re piecework labor smoothing the rough edges of a machine that’s being marketed, explicitly, as the thing that will replace them.

And it gets worse, because of the loop’s cruelest feature: the human in the loop is training their own replacement. Every time a Weave specialist takes over the arms to fold a tricky bedsheet, that intervention becomes labeled data: a demonstration of exactly how to handle the case the robot failed. It’s the same flywheel the book traces in the bootstrap paradox and the tool that trains its replacement: the labor you contribute to make today’s system work is the fuel that makes tomorrow’s system not need you. Tesla was blunt about this logic years ago: it used millions of human drivers to train its self-driving AI. Now humanoid robots need someone to fold your laundry, on camera, a few thousand times, before they can do it alone. Guess who’s holding the shirt.

The Tell: Why Companies Hide the Human

If the human-in-the-loop stage is normal, predictable, even honorable engineering, why does everyone bury it in a privacy disclosure and a footnote?

Because “fully autonomous” is worth a fortune and “mostly autonomous, with a person on standby” is worth a Tuesday. Autonomy is the word that moves valuations, closes funding rounds, and lets a company book a human’s wage as a temporary cost instead of a permanent line item. A robotaxi with 1.5 staff per car is a labor-arbitrage business wearing a software company’s margins. A home robot with a remote operator is a staffing agency cosplaying as a hardware breakthrough. The theater isn’t vanity. It’s the business model.

There’s a subtler cost, too, and the book cares about it more than the accounting. When companies hide the operator, they corrupt the public’s map of reality: we collectively overestimate how far automation has come, right at the moment we most need a clear-eyed read to plan the transition. If you think the Labor Cliff is already here because a robot folded a shirt on stage, you’ll panic early. If you think it’s science fiction because “the robots are all fake, it’s just teleoperation,” you’ll be blindsided when the dimmer finally hits the top. Autonomy theater breaks the instruments right when we need to read the timeline honestly.

To be clear, this cuts against the hype and the dismissal. “It’s all just teleoperated puppets” is the lazy skeptic’s version of the same error. The remote operator is not proof that automation is fake; it’s proof that automation is underway and passing through the stage it always passes through. The shirt did get folded. The car did drive most of the miles. The dimmer is moving. It’s just not a switch, and anyone selling you a switch is selling you theater.

“But Teleoperation Is Just a Temporary Bridge”

Sometimes, yes. And when it is, wonderful: that’s automation working as advertised, a scaffolding that comes down once the building stands. The honest version of the story sounds like this: “Today a human catches 8 percent of cases; next year, 3 percent; eventually, near zero.” Great. Say that out loud, put the operator on the payroll and in the disclosure, and let the public calibrate.

But “temporary” is doing a lot of load-bearing work in that sentence, and it has a habit of never ending. Some tasks have a fat tail of weird edge cases (the sock stuck to a duvet, the toddler who wandered into frame, the construction cone in the wrong lane) that resist automation long after the demo looks finished. For those tasks, the human-in-the-loop isn’t a bridge to the far shore. It is the far shore. And a bridge you quietly staff forever, while telling investors it’s coming down next quarter, is not a bridge. It’s a hidden workforce with a good PR department.

The book’s stance is not “teleoperation is bad.” It’s that whether the loop is temporary or permanent is a fact about the world we deserve to know, because entire policy questions (how fast the Labor Cliff arrives, who needs a floor, who’s owed recognition) depend on the answer. Hiding it doesn’t just mislead investors. It sabotages our ability to plan the most important economic transition in human history.

The Unscarcity Read

Strip away the theater and ghost work delivers a gift: it shows you, in miniature, the exact failure mode the whole framework is built to prevent. Here is a form of labor that is essential, invisible, precarious, uncredited, and busy training its own obsolescence. That’s not a glitch in the automation economy. Left alone, it’s the destination.

The book’s answer runs on two rails. First, the Foundation: an unconditional floor (housing, food, healthcare, energy, compute) that doesn’t ask whether a machine has finished replacing you before it guarantees you can live. The tragedy of the ghost worker is that they’re caught in the gap, too automated to be a “real” job, too human to be automated away, and falling through a floor that doesn’t exist. Build the floor and the gap stops being a trap. Second, Impact: a system that makes contribution visible and recognized instead of hidden and disposable. The whole pathology of ghost work is that the labor is deliberately erased from the record. A civilization that runs on legible contribution, where Civic Service and real work earn standing you can see, is the structural opposite of a business model that depends on pretending its workers aren’t there.

There’s a cousin to this problem worth naming: the debate over where AI agents must pause for a human isn’t only about hidden manual labor but about which decisions we should never fully delegate, a question the book takes up in Human-in-the-Loop: Where AI Agents Must Stop. One asks who’s secretly doing the work; the other asks who should deliberately stay in the loop. Both land on the same truth: the human doesn’t vanish from the system. The only choice is whether we’re honest about where they stand.

So the next time a robot folds a shirt on stage, ask the one question the theater is built to suppress: who else is in the room? Maybe no one: maybe the dimmer really did reach the top for that task, and the Frontier of what’s automatable just moved. Or maybe there’s a person in another country with a headset and a joystick, folding your laundry through a machine, being erased from the story in real time.

Knowing which is not a technicality. It’s the difference between planning the transition and being ambushed by it. Unscarcity is the blueprint for making sure that when the ghosts finally step out from behind the curtain, there’s a floor under their feet and a name on their work.

Sources

• Weave Robotics’ Isaac robot folds laundry with remote human help — New Atlas (Isaac 1, $7,999/$449-mo, teleoperation, camera/privacy disclosure, California deliveries late 2026)
• Is Amazon’s ‘Just Walk Out’ powered by AI or by workers in India? — KUOW/NPR (1,000+ reviewers in India; Amazon’s rebuttal)
• Cruise confirms robotaxis rely on human assistance every four to five miles — CNBC (2–4% remote assistance; ~1.5 staff per car)
• Tesla Optimus bots were controlled by humans during the ‘We, Robot’ event — TechCrunch (teleoperated crowd interactions)
• Mary L. Gray & Siddharth Suri, Ghost Work (2019) — the foundational account of hidden human labor behind “automated” systems

Related Articles

• The 2025-2030 Labor Cliff — The real edge of automation, and why the dimmer matters more than the switch.
• The Humanoid Robot Revolution — The body of the revolution: demo autonomy versus durable autonomy.
• The Tool That Trains Its Replacement — When your everyday work becomes unpaid training for your own automation.
• The Bootstrap Paradox — The flywheel that turns today’s human labor into tomorrow’s machine.
• Human-in-the-Loop: Where AI Agents Must Stop — The governance cousin: which decisions humans should never delegate.
• Employment Statistics — The numbers behind the anxiety, and the jobs that don’t show up in them.
• The Foundation — The unconditional floor that closes the gap ghost workers fall through.
• Impact — Making contribution visible and recognized instead of hidden and disposable.