Nikita Rudin spent his doctorate at ETH Zurich building simulation tools that much of the robotics industry now trains on, and he thinks the industry is quietly buying its way around the hard problem. Writing in The Robot Report, the Flexion co-founder argues that humanoid robotics has grown dependent on teleoperation, and that a machine which cannot handle anything new without fresh human input is closer to a labor system than an autonomous one. He is right about the economics. The general rule he draws from them does not hold everywhere, and hard-tech founders who miss the difference will pay for it.
What Rudin is claiming
Over the past eighteen months, he writes, humanoid robotics companies have raised billions of dollars, and most of that money is quietly paying humans to operate robots. Teleoperation datasets, by his figure, are more than 100,000 times smaller than the corpora behind language and vision models. A commercial ecosystem has grown to fill the gap, with startups in China, India, Europe and the United States selling teleoperation data and recruiting operators mostly in lower-wage economies. His alternative is reinforcement learning in simulation, which scales with compute rather than headcount. Flexion raised $50 million from DST and NVentures to build it.
Founders should take that economic argument seriously. If your training data cost scales with human hours, then every improvement in the product costs you more labor rather than less, and that curve does not bend on its own.
Where the general rule breaks
Rudin's framing treats a need for fresh human input as evidence of failure, and that does not travel across hard-tech. I spent years in surgical robotics, where the human in the loop is the product rather than a stage of development. Laparoscopic surgery puts four or five rods into a patient, and da Vinci won because it was built to control more instruments than a surgeon has hands. Remove the surgeon and you have removed what hospitals are buying. The Galen design goal ran the same direction: sit smaller than the assistant it replaced, keep the surgeon in their normal working position, make the hard part of the procedure disappear under their own hands. Nobody in that operating room is asking for autonomy. The useful question is not whether a human sits in the loop, but whether that human is there by design or by accident.
How to tell which one you have
Capital is arriving faster than reliability. Manufacturing Dive reports Barclays figures putting humanoid installations at roughly 2,000 units in 2024, 15,000 in 2025, and a projected 60,000 this year, while Crunchbase counts $18.8 billion raised by robotics startups in the first half of 2026 against $15 billion for all of 2025. In the same piece, GM's Mikell Taylor sets the operating bar: 80% success is failure in production. Teleoperators live in the space between those numbers.
The test I would apply is uncomfortable on purpose. Point to the human on your org chart and your bill of materials, priced and staffed for the next five years, and defend that as the product you meant to build. If the operator turns up in the demo but never in the unit economics, you are describing an unfunded liability to investors and calling it a roadmap. That risk sits heaviest on warehouse and industrial robotics founders raising on autonomy claims, and on physical-AI teams whose pilot deployments quietly run hybrid. For an interventional or surgical device the answer inverts, and the human in the loop belongs in the pitch as a feature you charge for.
Humanoids give really good demo, and that has always been the tell. I do not much care whether a founder keeps a person in the loop. I care whether they can say out loud which version they are selling, because the ones who cannot tend to sell autonomy to investors and staff it with contractors. Pick the honest version and price it, and at least the argument becomes about engineering instead of story.
From Dave’s video library
In this one I look at where AI agents already make decisions that affect people, and why the real risk is handing a limited system the keys with no human override.
I’m here to help you scale.
Work With DavePrefer a smaller first step? Book a $500 one-hour working session →
Dave Saunders is the founder of Base Reality Group and a Fractional CPO for hard-tech founders. He was a founder and operator at Galen Robotics, where the surgical-robotics platform earned FDA De Novo authorization in 2023, and he managed a 35-patent portfolio licensed from Johns Hopkins. He wrote Founders Who Finish and publishes The Build. More about Dave →