Robot startups that won by selling hourly rates, not AI
When Tutor pitches to first-time robot buyers, it never mentions models or AI—just one number: $16 an hour. The strategy reflects a vast market gap: America deploys only 38,000 industrial robots annually, leaving 90% of factories untouched.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
Start with robot deployment, not model training
Josh observes that 99% of robotics startups begin with a model or data insight, but few ask what a robot actually needs to do. Tutor inverts this. Even if perfect software could connect 8 billion humans' physical intelligence to any machine, robots still wouldn't exist, supply chains wouldn't exist, and customers wouldn't exist—because robotics remains a cottage industry even in manufacturing and logistics. So Tutor deploys robots in the real world first, then builds datasets, models, and supply chains from that experience.
— JoshThere's no internet for physical labor, so build one
Josh explains the company name by contrasting social media and email—natural human behaviors that inadvertently created a training dataset for LLMs—with physical work, which leaves no digital record. When Josh picks up his coffee cup, no one records the pressure or angle his fingers used. Tutor's solution isn't to invent physical reasoning from scratch. Instead, it deploys robots into the real world where human remote operators (the "tutors") teach machines new tasks. This amounts to building an internet for robot training data, and the process itself generates economic value by making customer goods cheaper.
— JoshAmerica deployed 38,000 industrial robots last year. That's it.
Josh offers a sobering comparison: U.S. factories and warehouses took delivery of just 38,000 industrial robots in 2025—a volume so small that even one successful humanoid robot model would barely register statistically. The problem runs deeper than quantity. Robotics OEMs like Fanuc and ABB make machines either overseas or for systems integrators handling fast-repeat tasks—not designed around the choices needed to work with advanced LLMs. Tutor had no choice but to design and manufacture its own hardware rather than source off the shelf.
— JoshTwo robots handle 80–90% of warehouse and factory jobs
Tutor makes only two models. Cassie is a single-arm robot with a suction gripper—a "workhorse" powered by a foundation model with 200M–400M parameters. It handles heavy pallets and cartons, solving the labor problem of turnover (workers average 55 years old and quit within three months). Sunny is a wheeled (not legged, because factory floors are polished) industrial humanoid robot engineered to run continuously for three to ten years at human speed and reliability, with industrial-grade collaborative safety. Josh claims these two robots cover 80 to 90 percent of manufacturing and logistics work.
— JoshThe sales call opens with an hourly rate, not a model
Tutor's customers are almost entirely first-time robot buyers—people familiar with older automation who assume robots can't apply to their business. Breaking through means abandoning the pitch about AI, models, or tutors. Instead, Tutor calls and asks, "Have you considered a robot at $16 an hour?"—a proposition they've never heard. This alignment of incentives lets Tutor move multiples faster than traditional industrial automation. First call to signature: 40–60 days. Signature to robot running on site: roughly 40 days. Installation often happens in a day or even hours.
— JoshBetting on a big customer and custom work is how robot startups die
Josh has observed this pattern repeat five to ten times: a research-heavy robotics company lands an LOI from a large customer, spends two years building a hyper-customized system, and it either doesn't work or the customer's problem disappears. The company dies. His lesson: work with diverse small and medium customers. That constraint—far from being a weakness—forces the company to build a true general-purpose robotic body and platform, not a Rube Goldberg machine built for a single buyer.
— JoshDemand shifts entirely when factory-built homes undercut site-built by just 1 percent
Josh recalls an example from a Wisconsin housing factory worker: factory-built homes currently cost 5% more than site construction. But the instant factory construction becomes just 1% cheaper—even fractionally—developers will shift all demand to the factory model. He uses this to argue that robots don't unlock linearly declining costs. Cross a critical threshold and entire industries flip. The same logic applies to materials recovery at landfills and similar markets where cost inflection points trigger structural shifts in where work gets done.
— JoshChina's bet on yesterday's automation may be a long-term miscalculation
When pressed on how robotics will reshape factory location and regional manufacturing patterns, Josh pivots to geopolitics. China invested heavily in the previous generation of industrial automation, and it paid off handsomely. But over the long term, it may be a massive mistake. The U.S. is now undertaking greenfield automation using general-purpose robots without tearing out old lines first—an advantage China lacks. Josh acknowledges that supply chains remain deeply linked: Chinese and American firms both invest in each other's markets and build their own capabilities. Tutor itself is moving more component supply vertically back to its Watertown factory because the company has become one of the largest demand sources for certain OEMs.
— JoshIn their own words · checked verbatim
That network of human intelligence for LLMs and other kind of digital world models by natural emission has created this beautiful data set that we can use to train general purpose models.
Josh9:27
a human had to put that in a box, which is a crazy revelation for most people.
Josh21:57
We call them up and we're saying, hey, have you ever considered a robot for 16 bucks an hour? And they haven't, right?
Josh23:58
a robot is a robot unless it works, in which case it's a machine.
Josh25:58
anybody who's built products knows that you can't just one and done a product, right? Doing product requires iteration.
Josh30:10
we have this framework of the value of a robot is you take how much work a robot can do and you divide it by its cost. And that's it, right? Everything else is commentary.
Josh36:19
I mean, you're really competing with like something that you feed rice to and it's very, very efficient, right? Like the human body.
Josh37:19
this is going to be a great equalizer between the United States and other companies, or excuse me, other countries like China invested in the previous regime of industrial automation. And in some ways that made them very successful. And in other ways, I think in hindsight, that might end up being an enormous mistake.
Josh39:20
Figures
| Tutor headcount | ~160, primarily based in Boston | 2:07 |
| U.S. factories and warehouses without robots | Nine out of ten | 23:58 |
| Sunny robots at Data Factory One | 100 | 19:56 |
| Average time from first call to signature | 40–60 days | 24:58 |
| Average time from signature to robot on-site | ~40 days | 24:58 |
Glossary
- Teleoperator / Tutor
- A human remotely controlling a robot and teaching it new tasks; the interaction data simultaneously trains the model.
- Greenfield market
- A market with little to no prior competition or product penetration; essentially starting from scratch.
- Human-Robot Interaction (HRI)
- The academic field studying how humans and robots collaborate and build trust.
- Action diffusion transformer
- A model architecture that converts a model's general knowledge into specific sequences of robotic actions.
How to listen
Founders and investors interested in robot hardware deployment, manufacturing automation, SMB sales strategy, or U.S.–China supply-chain dynamics.
27:00–29:00 covers robot naming trivia with low information density; feel free to skip.