The Robots Already Working in American Factories | Tutor Intelligence

Tutor Intelligence bets the whole robot stack wins the factory

Tutor Intelligence sells robots that already do factory and warehouse work, and the data they generate is the point. Josh Gruenstein argues a robot company has to own the full stack rather than one layer, since that is what produces the training data and the manufacturing scale. Nine of ten American factories and warehouses have no robots, so the pitch is a robot for $16 an hour.

Takeaways

• Robotics is one stack, not a layer. Gruenstein says a robot company has to own the full stack, because dropping a layer eliminates 90% of the trade-off space. No robot startup has been massively economically successful because the intelligence primitive is missing and the market is disaggregated.
• Data from paying work is the real asset. Tutor's robots connect to human teleoperators it calls tutors, who teach new tasks and generate an internet of robot training data. Data gathered while robots do paying work is guaranteed valuable, Gruenstein says, unlike data from camera-wearing people sent into the field.
• Tutor builds its own robots on purpose. The US deployed only 38,000 industrial robots in 2025. Gruenstein says Fanuc and ABB machines are built for systems integrators and fast repetitive motion, not co-designed with frontier models, so Tutor designs and builds its own robots in Watertown, Massachusetts.
• The pitch is $16 an hour per robot. Nine of ten American factories and warehouses have no robots. Average time from first call to signed contract is 40 to 60 days, from contract to robots working on site about 40 days, and robots stand up in production within a day.
• Mid-market customers force a general product. Most US manufacturing and logistics businesses are small or mid-market. Gruenstein says a diverse customer base is a forcing function for building a general robot, unlike startups that build one custom system for one large customer.
• Value is work done divided by cost. More of Tutor's cost is ongoing inference and teleoperation than upfront capital expense. Gruenstein's framework is that a robot's value is the work it can do divided by its cost, and a customer can save a million dollars a year without capital outlay.

What comes next

• Cassie and Sunny can handle 80% to 90% of the work inside manufacturing and logistics, but cannot go outside. Tutor's 100-robot Data Factory One in Boston collects real-world data, and legged robots would be needed for construction and landfills.
• Gruenstein predicts general-purpose robots will be a great equalizer between the US and China, because the US is doing greenfield automated industrial builds while China must tear down its previous automation regime.