1X’s Home Robot Is Open for Preorders — and for Remote Operators

When 1X opened preorders this week for its $20,000 humanoid home robot NEO (or $499/month), the company framed it as a historic moment: humanoids, finally, in real homes in 2026. US tabloids said “Americans can now order a real-life home helper” and quoted 1X’s founder talking about closing the gap between science fiction and reality (The Sun, 1X on YouTube, 1X on its site). Within hours, though, the story shifted. Viewers noticed that every smooth marketing clip looked like a VR session. MKBHD pointed on X that the “home robot” in the video was being piloted, not doing autonomous manipulation, and that this looked very close to Tesla’s data-collection strategy for FSD — ship early, learn on customers (MKBHD on X, Humanoids Daily). Reddit, Hacker News and even robotics subreddits found something else: 1X’s own career page is recruiting “Robot Operator” and “Software Engineer, Teleoperation,” explicitly saying operators will run robots “as the technology scales, in customer homes.” That is straight from the company, not from critics (1X careers, Robot operator job, teleop engineer).

So the real headline is not “humanoid finally here.” It’s “humanoid will come to your house, and for a while a person in Palo Alto will watch your kitchen through a camera.”

What 1X actually shipped

Every serious review landed on the same line: NEO cannot yet do household tasks on its own. The Wall Street Journal test put it plainly — “it’s still part human” — and described a 1X operator in VR controllers stepping in to load the dishwasher and pick up objects (WSJ, paywalled; summarized at Reddit). Engadget said owners will use an app to schedule when the remote operator can take over, and that teleoperation is how the robot will “learn chores” (Engadget). Road to VR stressed the same thing: the autonomous future is the promise, telepresence is the present (Road to VR). BGR basically told readers not to buy yet because of this gap (BGR).

This is exactly what the 1X job postings describe: operators follow scripts, collect high-quality data, annotate sessions, and their work “feeds directly into the AI training process” so that future robots can do the same tasks on their own. That’s the data flywheel — sell the robot now, hire humans to make it look competent, train on those sessions, reduce human input later.

Why 1X is leaning so hard on VR

At first glance VR teleoperation feels like a step backwards — the whole point of a $20,000 humanoid is not to have someone remote-control it. But for 1X this is the plan, not the patch.

Teleop is the data hose. If you want a robot to learn how people actually unload dishwashers or fold T-shirts in real, messy homes, you need high-fidelity demonstrations. Headset + controllers gives you full-body, task-level motion capture at human speed. That’s far richer than clicking waypoints on a tablet. It’s exactly what their job posts say: “operate our robots in customer homes” and feed those sessions into training.
VR hides the mismatch between human dexterity and today’s actuators. NEO’s arms are tendon-driven, the whole robot is heavily padded, and the company cares a lot about safety around kids and pets. All of that makes the robot move like it’s in thick water. Putting the human “inside” the robot with VR makes those micro-hesitations look less like a software bug and more like a cautious human.
Low latency is a tech problem — but solved in labs, not suburban Wi-Fi. 1X talks about “minimal latency” and clearly has the streaming stack for showroom demos. But even in friendly reviews, viewers could spot the human-in-the-loop hesitation — the slow wrist rotation, the pause before placing objects. That’s what remote looks like once you leave the fibre-connected office. So yes, 1X does have “teleop as a capability”; no, it does not yet have “teleop that always feels natural in any customer home.”
It’s how you scale humans across robots. A slow, padded, safety-first robot can be supervised intermittently. That’s why this model is compatible with moving at least part of the work to cheaper locations later. You can imagine one operator in Guadalajara or Manila watching over several Californian robots — not today, maybe, but the architecture points there. And that’s exactly why people started calling it “Mechanical Turk with a $20K body.”

So the real question for 1X is not “why VR?” but “for how many years is VR part of the product?” If the answer is “until the models catch up,” then customers are being asked to co-fund that catch-up period — and to let strangers look into their homes while doing it.

Are they a robotics company or a remote-care company?

Right now, uncomfortably, they are both.

The product story: you buy a humanoid that can “do chores,” and when it gets stuck, a vetted 1X operator takes over. That’s a service business wearing a robot.
The investor story: 1X is building general-purpose embodied AI with its own actuators, its own data pipeline and its own models. That’s a robotics + AI platform story.

Those stories clash at exactly one point: privacy. A household service company is allowed to have humans in your space — cleaners, carers, dog-walkers. A robotics company is supposed to make the human disappear. 1X is trying to be both: “here’s a robot in your living room” and “also, when it’s confused, a person will see your living room.” That tension is what the internet smelled in 48 hours.

And it gets sharper when you look at scaling. The consumer-facing pages and the WSJ demo emphasise “scheduled,” “opt-in,” “vetted” operators. That’s how you sell to wealthy early adopters. But the scaling narrative — the one VCs want — is “we can run hundreds or thousands of robots from a central operations centre.” Those are not the same promise. You can’t have unconstrained centralised ops and boutique privacy guarantees at the same time unless you build very strong task-level encryption, logging, and revocation for homeowners.

“Isn’t this just ROS with good video?”

Probably not — and that’s the bit that gets lost when criticism goes straight to “scam.”

1X actually has some real robotics depth. The company lineage goes back to Halodi Robotics in Norway. They’ve been working on safe, lightweight actuators, on whole-body control, on getting a human-scale machine to co-exist with people without being caged. That’s not something you download from GitHub.

Where they are unabashedly slick is in the AI and marketing layers. They moved fast to align themselves with OpenAI-adjacent embodied work, then to show off “world-model-style” reasoning, and now to demo with NVIDIA’s GR00T N1. That’s the part that looks like “wrapping” because it sits on top of a still-maturing autonomy stack and makes the whole thing look more 2026-ready than it really is.

So a fairer read is: this is a competent mechatronics team that realised pure autonomy wasn’t going to get into homes in time, so they stapled on human-in-the-loop, and then told the bigger AI story to pay for it. That’s different from “they have no tech.”

The marketing vs. reality gap

This is not new in robotics. The industry has a habit of showing the best run — the one that took 20 takes — because it is trying to raise money in a space where the hardware is expensive and the autonomy is brittle. Boston Dynamics spent years publishing parkour videos; Tesla’s 2025 Optimus demo focused on precision grasping, not on all the times the robot missed; Figure AI keeps showing carefully staged kitchen and warehouse tasks; even Agility Robotics, which is probably the most honest of the bunch, demos Digit doing pick-and-place for Amazon, not Digit tripping over in a messy garage (Business Insider on Tesla Optimus, Humanoids Daily on Figure vs. 1X, Agility fundraise).

What made 1X’s launch feel different is that the teleoperation was so visible people could spot it from home. MKBHD and others froze frames and saw the classic human-in-the-loop tells — slow wrist rotation, micro-pauses before placing objects, wandering gaze — all classic VR teleop signatures. And then the internet found the operator jobs. So the usual “we’re showing our best” spin didn’t land.

The Tesla / Waymo comparison needs fixing

A lot of people (including 1X’s own fans) jumped straight to the Tesla FSD comparison: “they’re shipping before it’s autonomous so they can learn on customers.” That’s only half right.

Tesla uses customer miles to improve a system that runs on public roads. The sensitive thing is your driving, not your bedroom.
Waymo/Cruise do have remote staff, but they don’t tele-drive your car the way 1X tele-drives your robot. They provide high-level assistance for edge cases — “yes, you can go around this cone,” “take the second exit” — because full tele-driving at road speeds would be unsafe and too latency-sensitive. Also, crucially, they are watching public space, not your child’s playroom.
Baidu’s Apollo / Robobus / Robotaxi is closer to 1X in spirit: there is a documented path for remote operators to take over difficult scenarios. But again: public roads, public cameras.

1X moved the human-in-the-loop model inside private space. That’s the real leap, and it’s why the privacy conversation dwarfed the cool factor of “humanoid in your house.”

So the better analogy is: this is like Cruise’s “remote assist” plus Tesla’s “learn on customers,” but strapped to a mobile camera with arms that lives in your living room. That’s a very different social contract from “your car occasionally phones home.”

The cheap-labor question

Once people saw “remote operator” in the job listings, the internet did what it always does: “so I’m paying $20,000 to hire a hidden worker to pick up my socks?” That’s not what 1X is saying — today’s listings are for US-based operators at roughly $22–$32/hour and for engineers building low-latency VR control. But the model is clearly compatible with moving some of this work offshore as long as latency and time zones allow it. If your core promise is “a human will take over when the robot is confused,” there will always be pressure to make that human cheaper. That’s not a scandal; it’s just how ops businesses evolve. But it does make the “humanoid breakthrough” headline look a bit thinner.

The 2026 problem

1X says 2026 for deliveries. It doesn’t say Q1 2026 or “first half” or even “late 2026.” It just says 2026. That’s a tell. When hardware companies are very confident, they give months. When they give only a year, it usually means “we know we can get something into some homes, but we don’t know how many or what feature level.”

What has to happen between now and “real” 2026 shipments is non-trivial:

they need to make teleop low-latency in real residential networks
they need to hire and train enough operators to cover early customers
they need to collect enough demonstration data in diverse homes to start automating the common chores
they need to convince regulators and consumer groups that a remote human watching you fold laundry is not a surveillance product

Any of those can slip. And 1X does not get to slip alone: Tesla wants Optimus in factories in 2025 and into some customers by 2026; Figure AI is telling Dreamforce it is “building a new species” and raising hundreds of millions to do it; Agility just raised about $400m to scale Digit for logistics. If the others manage even semi-autonomous pilots in the same window, 1X’s “humanoid in every home” story will look more like “telepresence maid.”

Robots, like cars, need to survive the average day

The link with autonomous driving is straightforward. Waymo, Cruise, Tesla — all of them had perfect-looking demo videos before they had reliable services. The hard part is not the video; it’s handling the 5% of weird cases without sending a human to take over. Robotics is now in the same place: the manipulation demo is impressive, but the long tail of home environments is brutal. People leave socks on open dishwashers. Light changes. Kids run through the room. Dogs bark at cameras.

1X’s answer is “we will have humans to cover that until the AI learns.” Figure’s answer, made explicit this week by Brett Adcock, is “we will not teleoperate; we will eat the autonomy pain now because we don’t want the privacy and labor complexity later.” Both are rational. But only the 1X answer is sellable to consumers right now — which is also why it is the one getting roasted.

What 1X still has to answer

How long is teleop first-class? “We will unlock more abilities over time” is fine, but 2026 buyers deserve to know whether “human in the loop” is a 6-month bridge or a 3-year reality.
Will operators stay in high-wage jurisdictions? Today it’s Palo Alto/Norway. Tomorrow it could be Mexico or the Philippines. If that happens, what happens to privacy, data retention, and legal recourse?
Will customers get per-task logs and recordings? If the product is “a robot with occasional remote eyes,” the only respectable way to do it is to let households see exactly when those eyes were on.
What is the average-case autonomy today? Not the hero dishwasher video. The median home. The dog. The toys on the floor. Just like AVs, robotics needs to get used to publishing the boring numbers.
What happens if 2026 ships with teleop still dominant? Does the price come down? Or do early adopters end up paying for the human labour behind the scenes?

Will spring come — and for whom?

The good news for the sector is that 1X just proved there is at least a small market of people who will put down $20,000 or $499/month for a humanoid even if a human is secretly driving it. That is a real, paying testbed. It will generate the kind of messy, non-lab data that robotics teams have been begging for.

The bad news is that this model is expensive to run and easy to mock. If the company can’t automate fast enough, the unit economics look a lot like remote labour in a robot suit. If it automates too aggressively, it will hit the same failure cliffs that made self-driving slow down. And all the while, better-funded rivals with different philosophies — Tesla with factory-first Optimus, Figure with “no teleop,” Agility with non-humanoid logistics, Apptronik with industrial pilots — will keep shipping demos that look cleaner because they don’t have to expose the human in the loop.

So will there be a spring? Probably. But, like AVs, spring will belong to the companies that can keep paying through years of human-in-the-loop operations until the autonomy really is good. 1X just bet that consumers will help fund that phase. The question is whether they will still be around, and still trusted, by the time the robot can actually empty a dishwasher without someone in a VR headset watching.