The Clankers Have Left the Lab

Somewhere between the espresso machines and the exhibit booths at Paris' Machina conference — a gathering billed, without irony, as "AI in the real world" — Jason Calacanis found the thread that stitched four separate robotics companies into a single, slightly unnerving story. It isn't the walking. It isn't the backflips. "They don't care about the robot," Peter Fankhauser told him. "Actually, they don't even want the robot. They want the data. The robot is a means to an end."

Watch on YouTube →

That reframing runs through the whole afternoon. For years, robotics has been a spectacle — a thing you show your friends, a dog that dances, a humanoid that gets kicked and heroically stands back up. Calacanis, playing the wide-eyed sci-fi enthusiast who has clearly read too much industry news, kept trying to steer the conversation toward Terminators and ocean floors. His guests kept steering it back toward invoices.

Act One · The Dog That WonWhy four legs beat two

Fankhauser is the co-founder and CEO of ANYbotics, and his machine is the ANYmal — "a lot of you have puns," Calacanis noted, correctly. The company is ten years old: five years as a research lab, five as a business that sells inspection to people who own things that explode. Nearly twenty years into the four-legged form factor, Fankhauser has a clean answer for why the "dog" got to real-world scale before the humanoids did.

"In nature, a lot of animals have four legs, so there's a reason to that," he said. Stability. Footholds. A wide stance that holds on in slippery, snowy, rain-slicked places where a two-legged robot would face-plant and, in Calacanis' vivid phrasing, "break somebody's ankle." When the host floated putting four legs on a humanoid — inventing, live, the "centaur" cafe robot he insisted "should be the new standard" — Fankhauser gently agreed it was possible and moved on.

The economics are brutal and simple. These robots cost low hundreds of thousands of dollars, plus service contracts. But the assets they inspect — pipelines, electric arc furnaces, offshore transformer stations — lose "hundreds of thousands per hour" when they stop. So a robot packed with thermal cameras, acoustic microphones and gas sensors that catches a micro-leak before it becomes downtime pays for itself in minutes. Some customers run these missions 40 times a day, up from once or twice by hand. Offshore, where every helicopter flight for a human inspector runs into the tens of thousands, the math gets even friendlier.

ANYbotics even built a robot certified not to throw a spark, so it can walk into methane-heavy explosive atmospheres — the Permian-basin nightmare where, as Calacanis put it, "people seriously die." That, everyone agreed, is the whole point. Send the machine where the human shouldn't be.

Act Two · The Home Humanoid"Put the hat on Neo. I am Neo."

If Fankhauser is the pragmatist, Bernt Bornich, founder and CEO of 1X, is the true believer with a shipping deadline. His Neo is a household humanoid, pre-sold to early adopters who were promised delivery in 2026. Would he hit it? "You got to keep your promises," he said. "So, we will ship in 2026." A handful of customers, slowly, rough around the edges. "They're going to fall," he admitted, cheerfully.

The real news he "dripped in": Neo is becoming a platform. An app store for skills — if you want your robot to make a salad, some hacker builds the salad skill and you subscribe. 1X sold out its first 10,000 pre-orders in days; the subscription lands somewhere around $500 a month, the Tesla-deposit model for people Calacanis called "the Vanguard, the earliest of the early adopters."

But Bornich's deepest bet is stranger and more interesting than any app store. It's about data — the "catch-22" of robotics. Language models trained on the entire internet; there is no equivalent trove of torque commands and tactile forces for robot control. So 1X made a decade-long wager: build a robot human-like enough that it can learn from all the video of humans that already exists.

That's why 1X obsesses over how much a fingertip deforms, the friction of skin, the impact energy of a hand meeting a table. Get the robot close enough to human, and all of YouTube becomes training data. Teleoperation doesn't disappear in this world — it becomes "expert in place," the ability to beam a surgeon, or a CEO, anywhere. Bornich's own favorite use case is running his company from the road: "Put the hat on Neo. I am Neo. And that's actually pretty magical."

Then Calacanis asked when robots start building robots. Bornich didn't flinch. "I am extremely sure that we're less than a decade away from hard takeoff" — robots building robots, data centers, chip fabs, doing the mining and refining. "My current bet would be three years." He clarified, with a grin, that it's hard takeoff, not hard takeover. "We're going to do it right."

Act Three · The OGBoston Dynamics grows up

Amanda McMaster — "just McMaster, no McMasters" — is the interim CEO of Boston Dynamics, the company whose robots you've watched do kung fu and get shoved around for a decade. Calacanis recited its dizzying ownership history (independent, then Google, then SoftBank, now Hyundai) and got a clean "you nailed it" in return. Her message: the lab experiment is over.

"It's long past dancing at this point," she said. "It's now doing real work." Spot, the quadruped, has over 500 customers across 46 countries and logs more than 3,000 hours between human interventions — the most-deployed mobile autonomous robot on the planet, she claims. Atlas, the humanoid, will swap its own batteries by rotating its torso. The framing is relentlessly about outcomes: find a $3-million-a-day air leak, and nobody asks whether the robot dances.

On China, McMaster was blunt in a way her European counterpart was not. Should the U.S. allow Chinese humanoids at all? "No." Why? "It's not safe" — she cited quadruped data being "back-channelled back to China" and invoked the semiconductor cautionary tale. "We have to make sure that the rest of the world uses our platform rather than China's."

And weapons? Boston Dynamics holds an anti-weaponization stance, does explosive-ordnance disposal for the government, and draws the line at "Terminator robots." When Calacanis pushed — China is building armed robots, so aren't you obligated to? — McMaster held. "That's a tough question and I think we're going to have to answer it when the time comes." Calacanis, half-playful, predicted the answer would arrive the day President Trump called. She kept her focus where the money is.

Act Four · The ProfessorA snowball, not a singularity

Jonathan Hurst, co-founder and chief robot officer at Agility Robotics — and still a robotics professor — has been at this over 20 years, long enough to have lived through the "AI winters." He gave the afternoon its cleanest thesis: "It's very easy to make a robot that looks like a person. It's very hard to make a robot that can do useful things in human spaces. And we're starting to see that today."

What changed? Perception. Large language models handed robots the world for free — point one at a table and it knows that's a phone, that's paper, that's tea. Four years ago, you'd have hand-coded every object. But Hurst is a splash of cold water on the singularity crowd. There's no silver bullet, no magic model that solves robotics. "Think of it more like a snowball picking up steam going down a hill." It snowballs because people keep pouring in money and engineering, not because a switch flips.

His company's robot, Digit, hauls totes and bins in warehouses today. The milestone he's watching: Digit V5, out later this year, is the first humanoid that can step out of a work cell with no physical barrier between robot and human — the "scaling moment," because it took a bottom-to-top redesign to make a machine Amazon would actually let near its people. He gently reminded everyone he'd built a package-delivery Digit that climbed a customer's porch stairs some seven years ago. It's on the roadmap; it just isn't the best first market.

Asked why not build a General Grievous with four or six arms, Hurst reached for first principles: one arm can't lift big things, two can, a third is hard to justify, and beyond that "there's a lot to coordinate." Everything, he insisted, has to be "driven by a real need." His advice to anxious young people was the most human note of the day. Robotics is "a massive opportunity" — PhD track or blue-collar, "robot operators, assembling and building robots," even, only half-jokingly, "clanker maintenance." His hope: that our kids look back on today's dull, dirty, dangerous jobs the way we look back on coal miners in the 1900s and think, "I can't believe people did that work."

Four companies, four form factors, one quiet consensus. The robots are out of the lab. Whether the "hard takeoff" arrives in three years or ten, everyone on that Paris stage agreed it's coming in years, not decades — and that the machines will be judged, first and last, not by how they dance but by the invoice they justify. Skate to where the puck is going, as Hurst put it. The puck, it turns out, has four legs and a full charge.