THE ROBOTS ARE HERE: But Who’s Really in Control?

There’s something familiar about watching a humanoid robot move. Maybe it’s how easily we project ourselves onto it; the gestures, the rhythm, the faint illusion of empathy. For decades, we imagined these machines as servants or partners, mechanical mirrors of our humanity. But watching 1X’s Neo this week, I kept asking a different question: Who’s really moving?

The Wall Street Journal demo of Neo was mesmerizing. The robot, built by Norway’s 1X Technologies, is sleek, quiet, and almost human in its proportions. At 66 pounds, powered by tendon-driven motors, Neo pours water, loads a dishwasher, and navigates a living room with ease. It feels historic, until you realize that every motion is controlled by a human wearing a VR headset somewhere else in the world.

Neo by 1X: A Marvel of Form, a Mirror of Function

Neo’s engineering deserves credit. It is light, efficient, and safer than most of its peers. Its hand strength matches a human’s, and its structure includes built-in safeguards that prevent dangerous actions. In terms of design, this is one of the most advanced examples of human-safe robotics we’ve seen.

But the illusion of autonomy falls apart quickly. 1X admits Neo is not truly “intelligent.” The so-called AI is a human operator, a pilot moving its limbs through a VR interface and feeding the system data for training. Every cup lifted and every plate washed teaches the company’s neural network how to perform the same actions one day without supervision.

When Neo completes a task, you’re not watching artificial intelligence. You’re watching someone else’s hands, relayed through steel and silicone.

The Hidden Hand Behind the Revolution

That’s what makes this launch both fascinating and unsettling. The story is about the future of autonomy, yet the technology depends on one of the oldest concepts in robotics: remote control.

So why now? Why release a robot that still needs a human behind the curtain?

Because the data matters more than the product. Neo is a $20,000 household experiment meant to collect thousands of hours of teleoperated behavior. Each hour teaches its AI what human actions look like in the real world. Over time, the company hopes to build a system that no longer needs human pilots.

It’s a clever, if ethically complex, model. Early buyers are not just customers. They are data donors, allowing their homes to become live training grounds for Neo’s evolution. Privacy is “managed” by scheduled operation sessions, but the trade-off is clear: you give the robot access to your environment so that it can learn to be autonomous tomorrow.

The Wider Field: Who Else Is Building Our Mechanical Counterparts?

Neo is not alone. The humanoid race is accelerating, and each company is chasing the same dream in a slightly different way.

Tesla’s Optimus: The Hype and the Hand Behind It

Tesla’s Optimus keeps making headlines for folding laundry, walking unaided, and handing out candy in Times Square. The hardware looks solid, and Elon Musk says mass production is coming soon.

But some of the famous demonstrations were not as autonomous as they seemed. Tesla engineers confirmed that parts of the viral hand-dexterity demo, including the robot catching two tennis balls were teleoperated. Someone backstage was literally driving the performance.

Tesla’s long-term plan is to train Optimus using the same vision-based neural networks that power its cars. The company claims to have moved away from full teleoperation, but it’s still learning from guided behavior. Optimus is advancing, but not yet acting independently.

Figure 03: Autonomy First, Data Later

Figure AI has taken the opposite route. CEO Brett Adcock insists the company will not teleoperate its robots, relying instead on simulation and carefully supervised real-world data.

The Figure 03 robot can fold clothes, load dishwashers, and perform simple home tasks. It struggles, however, with the unpredictability of human spaces. Figure’s autonomy-first strategy avoids the privacy trade-offs of 1X’s teleoperation model but collects less data. The pace may be slower, though the approach is more transparent.

Agility Robotics’ Digit: The Workhorse

In Oregon, Agility Robotics has quietly deployed its humanoid Digit in commercial environments. Digit doesn’t pretend to be a house companion. It moves warehouse bins and assists with logistics. There’s no teleoperation here, just structured autonomy inside controlled environments. Less glamorous, perhaps, but far more practical today.

Unitree and the Eastern Contenders

China’s Unitree Robotics continues to surprise with its H2 and G1 robots, capable of precise, almost theatrical motion. Their public demos often feature lifelike dance or combat routines, likely guided by human teleoperators or scripted sequences. They are technically stunning, but whether they can act independently outside choreographed settings is still unclear.

Hardware Leaps, Software Lags

The hardware progress is remarkable. Lighter limbs, tendon systems, and efficient motors have made humanoids safer and smoother than ever before.

But autonomy is still the bottleneck. Fine motor skills such as grasping, balancing, adapting  remain an enormous challenge. Robots can perform defined tasks in predictable spaces, yet stumble in the messy reality of daily life.

Engineers sometimes call this the “robotics slop” phase: the awkward in-between where robots almost match human dexterity but not quite. They can pour water, but spill half of it. They can fold laundry, but can’t tell one fabric from another.

1X, Tesla, and Figure all agree on one thing: progress depends on data, and that means keeping humans in the loop for now.

The Ethics of Training a Machine by Living With It

The idea of being watched by technology isn’t new, but Neo introduces a new dimension - the physical one. Unlike a phone or smart speaker, it’s a robot that moves, observes, and learns from your body language.

1X says privacy safeguards are in place, yet the social contract is changing. Robots like Neo are not just tools; they are students. They’re learning our behaviors, gestures, and routines. We are teaching them how to act human.

The optimistic view is that this will enable humanoids to serve in homes, hospitals, and workplaces, improving safety and care. The darker view is that early buyers are unknowingly volunteering to be part of a massive data experiment that will outlive them.

The Year of the Teleoperated “AI”

If 2023 was the year of generative AI and 2024 the year of agentic software, 2025 might be remembered as the year of teleoperated intelligence. Many of these systems are not truly autonomous, but extensions of human control wearing the mask of independence.

And perhaps that’s fine for now. Every field passes through a stage where imitation fuels innovation. Robotics is no different. The important question is not whether these robots can act alone, but how responsibly companies handle the transition once they can.

So, Are the Robots Here?

In one sense, yes. Physically, humanoids are real, tangible, and improving at astonishing speed. But cognitively, they are still far from independence.

Neo, Optimus, and Figure 03 are not yet helpers; they are apprentices. Each one reflects human input, human pilots, human mistakes, human ambition. When companies claim the robots are here, what they often mean is that humans are still behind the curtain, shaping the illusion of autonomy.

That might be the most revealing truth of all. We are not only building robots; we are teaching them through ourselves. Every movement, every hesitation, every mistake becomes part of their memory. We are programming the future through our behavior.

The humanoids have arrived, but the real revolution is still learning how to walk. The question is not when they’ll take over, but whether we’ll recognize ourselves in what they become.