Five humanoid robotics companies are advertising jobs right now that share the same unusual requirement: wear a VR headset, sit in an office somewhere, and drive a $20,000 machine running around inside a stranger's home. Pay runs $22 to $55 an hour depending on where the operator sits. The robot does the loading. The operator does the thinking.
Those job listings are the finding. The reason it matters — that the fully loaded cost of remote human operation has fallen 65 percent in two years, from $340 an hour in early 2024 to $118 an hour by March 2026, according to the SVRC State of Robotics 2026 benchmark — is the explanation, not the news.
The news is that all five companies are still doing it, and the economics now work at scale.
1X, the Norwegian company shipping the NEO home robot at $20,000 or $499 per month, has made the arrangement explicit in its own job postings. Its careers page advertises Robot Operator roles, with operators running NEO robots in customer homes as the platform scales, feeding those sessions directly into AI training pipelines. The Wall Street Journal tested NEO and described exactly that: a 1X operator stepping in with VR controllers to load a dishwasher. A reviewer on X called out the VR controllers within hours of the launch video. Expert Mode, as 1X brands it, is teleoperation built into the product — disclosed, deliberate, and positioned as a feature, per Labellerr's review of teleoperation companies.
That transparency has a cost. Reddit and Hacker News found the postings immediately. The Wall Street Journal ran "it's still part human" as its headline. A competitor that does not disclose teleoperation looks more impressive in comparison, at least until the next clip surfaces and someone notices.
Brett Adcock, CEO of Figure, has drawn that line hard. He has accused unnamed competitors — and 1X by name — of deceiving the public by passing teleoperated demos off as autonomous. He called UBTECH's promotional videos CGI and MindOn Tech's household task demo a replay of pre-recorded movements. When an analyst published a timing analysis showing Figure 03 starting to turn before a verbal command was given — suggesting a human operator in the half-second lead — Adcock called it "soy stuff" without addressing the specific claim, per Humanoids Daily. At the White House Fostering the Future Together summit alongside First Lady Melania Trump, Adcock said Figure 03 was "autonomous. No human was in the loop for this." Figure has not published sensor logs, actuator timestamps, or any independent technical evidence that would settle the question.
The teleoperation job market is not theoretical. Global operators now work for $22 to $55 per hour in India, the Philippines, and Eastern Europe, or $65 to $120 per hour in the United States for domain experts with specific hardware certification. Tesla has advertised Data Collection Operator roles paying $25.25 to $48 an hour, requiring workers in motion-capture suits and VR headsets to train Optimus — seven hours a day or more. The SVRC benchmark shows twelve commercial humanoid platforms available for purchase or structured lease as of early 2026, up from three in 2024. The global robotics market hit $38 billion, up 34 percent year over year — the fastest growth in a decade.
When teleoperation costs one-fifth of what it did in 2024, a company can keep a human operator in the loop, run the robot in a customer's home, and still make the unit economics work — especially if one operator manages multiple robots as the software improves. That is a plausible path to full autonomy. It is also not what the demos show.
The question of whether a human is running the robot is not philosophical. It is a product disclosure question with safety implications, privacy implications, and investment implications that the market has not yet forced anyone to answer clearly. The companies that are honest about the human in the loop are being penalized by the companies that are not. The next time you see a humanoid robot do something impressive in a promotional video, watch the timing. Someone might have told it what to do first.
