Picture a fast food kitchen at 11:40pm. Fryer baskets hissing, one employee working the window, order tickets stacking up faster than anyone can clear them. Now picture a machine doing the fryer work instead. Not a demo unit on a trade show floor with a PR handler standing nearby, but a machine that runs the whole shift, every shift, and doesn’t call out sick.
That is roughly the pitch behind Atoms, Travis Kalanick’s new company, announced March 13, 2026. The stated focus is robotics for food service, mining, and transport, using AI vision to handle complex industrial tasks. The company frames its goal as automating physical work to bring about a “golden age” of abundance.
And yet the question following Kalanick around isn’t about fryers or mine shafts. It’s about robotaxis. Which tells you something about how people process news in this space.
Why the robotaxi question won’t go away
Kalanick founded Uber. Atoms lists transport among its focus areas. Kalanick has also been publicly analyzing autonomous driving, including commentary comparing Waymo and Tesla approaches to vision AI and scale, shared via a Twitter post by Sawyer Merritt on March 17, 2026, where he argued other players lack the necessary capabilities yet.
Add those three things together and the internet arrives at “robotaxi” in about four seconds. It’s a tidy story. It’s also an assumption stacked on top of an inference.
“Transport” covers an enormous amount of ground. Moving pallets in a warehouse is transport. Hauling ore out of a pit is transport. Shuttling containers around a port is transport. None of those require solving urban driving, negotiating with regulators in forty jurisdictions, or convincing the public to get into a car with nobody in the front seat. Industrial transport is a materially easier problem with a customer who already has a budget line for it.
My read from what’s actually been stated: Atoms is a physical AI company that happens to include transport, and a founder with strong opinions about self-driving cars. Those are two facts sitting next to each other. Treating them as one fact is how tech coverage gets sloppy.
The timing argument is the interesting part
Set the robotaxi speculation aside and there’s a more useful question: why 2026? Three conditions get cited as tailwinds for this bet.
- AI vision has crossed a capability threshold for handling irregular objects.
- Labor costs in developed markets have risen sharply.
- Capital for physical AI is available.
The first one is the load-bearing item. Irregular objects have been the wall industrial robotics kept hitting. A robot arm that welds the same seam a million times is a solved problem from decades ago. A robot that picks up a chicken thigh, or a rock of unpredictable shape, or a bag that slumps differently every time, is a different category of hard. If vision models genuinely cleared that bar, a lot of previously unautomatable work moves into range.
The second and third are business conditions, not technical ones. Expensive labor makes the math work. Available capital makes the buildout possible. Neither one makes the robot function.
What I’d want to see before believing any of it
I review AI tools for a living, and the pattern I see most often is a capability threshold announced well before it’s been crossed in production. “Crossed a threshold” is doing a lot of work in that tailwind list. Crossed it in a lab? In a controlled pilot? At a real site, at real speed, with a real failure rate that somebody is willing to publish?
Food service is a brutal test environment. Grease, heat, humidity, cleaning cycles, and staff who will absolutely work around your machine if it slows them down. Mining is worse in different ways: dust, vibration, and no help nearby when something breaks. Both are honest choices for a robotics company, and neither is a place you go if you want an easy first win.
So here’s my actual position. The Atoms thesis is coherent, the sequencing of vision capability plus labor economics plus capital is a real argument rather than a vibe, and the founder has shipped operationally difficult things before. That earns attention.
It does not earn a robotaxi headline. What exists right now is an announcement, a focus area list, and a stated ambition about abundance. Ambition statements are free. Uptime numbers cost something.
If Atoms ships a fryer robot that survives six months of unsupervised night shifts, that will be more meaningful than any autonomous vehicle announcement it could make. Boring, verifiable, and much harder to fake than a demo reel.
🕒 Published: