The $16-an-hour robot: most American factories don't have even one

In most American factories, there isn't a single robot. Tutor Intelligence's quote to these factories is: one robot, $16 per hour.

In US manufacturing, this price sits in a delicate spot. The federal minimum wage is $7.25, but once you factor in benefits, training, recruiting, and turnover costs, a production-line worker's true hourly cost is far more than that. $16 an hour buys a labor force that doesn't take leave, doesn't job-hop, and doesn't need health insurance. At least that's how Tutor sells it.

a16z's Martin Casado and Raghu Raghuram brought Tutor Intelligence CEO Josh Gruenstein on for nearly 45 minutes. Gruenstein's claim is that the robotics industry has had the problem backwards. Everyone stares at models, at humanoids, at general capabilities. Tutor's approach starts from the work itself: look at what jobs actually need doing in a factory, then put robots into them.

The key to this approach is in the name. Tutor. For work the robot can't do, a human tutor fills in. The tutor's nearby operations, corrections, and demonstrations all become training data. Tutor has built a data factory of 100 robots, named Cassie and Sonny. Each one works in a real environment, and each one generates data. Gruenstein says this is real-world reinforcement learning at scale.

This model has its clever aspects. A robot doesn't need to know how to do everything from the start. It only needs to do the work it can do while a human provides a backstop. The human tutor is not a temporary stopgap. It is part of the data flywheel. The more work gets done, the more data there is, the more capable the robot becomes, and the less the tutor needs to intervene. The cost structure changes along with it.

Tutor's customers are companies buying robots for the first time, with a focus on small and medium-sized businesses. This choice makes sense. Large companies have automation teams, integrators, and budgets for customization. SMEs don't. What they need is out-of-the-box, paid by the hour, and not counted as capital expenditure. The $16-an-hour price is aimed directly at labor costs, making the decision threshold much lower.

Gruenstein also talked about factories, China, and the US supply chain. Robot density in US manufacturing is astonishingly low; most factories don't have even one. This fact and Silicon Valley's daily talk about humanoid robots and artificial general intelligence feel like two parallel worlds. Smarter models cannot solve this problem. Tutor is betting on robots that are cheaper, more practical, and more willing to start with dirty, hard, and tedious work.

Progress in spatial reasoning is one variable. Frontier models have greatly accelerated robots' understanding of the physical world. But Gruenstein has not bet the farm on models. His logic is a systems problem: AI, hardware, manufacturing, and the work itself are all indispensable.

Whether the $16-an-hour price can hold depends on whether tutor costs can be diluted by data. If robots learn fast enough, demand for tutors falls, and the unit economics work. If they learn slowly, tutors become a hidden cost. Tutor's data factory is betting on the former.

Automation in American factories may not start with humanoid robots. It may start with a robotic arm that costs $16 an hour and has a tutor standing next to it. This picture isn't sexy, but the math works.

The full conversation is on a16z's channel: https://www.youtube.com/watch?v=lXkjXSul6yg