Robots Do the Work, Humans Back Them Up: Tutor Intelligence Raises $75 Million Series B

On November 2, a16z announced it had led Tutor Intelligence's $75 million Series B, with USV and Neo participating.

This company makes general-purpose robot hardware, and its deployment path differs from that of many peers. It doesn't wait for robots to be fully autonomous before putting them on the job. The robots first go into factories and warehouses to work, with remote human operators watching over them; these people are called "tutors." Humans guide and correct, while robots execute, and every operation accumulates training data for the underlying foundation model. The more it is used, the stronger the model becomes, and the more jobs it can take on.

a16z said in its announcement that Tutor's robots have already entered some of the largest manufacturing and logistics networks in the United States. Josh Gruenstein gave more specific numbers: the number of sites has nearly quadrupled from a year earlier; nearly one-third of this year's deployments came from repeat business from existing customers; and under contract plans, the number of robots will exceed 1,000 by the end of 2027.

Co-founders Josh Gruenstein and Alon Kosowsky met as undergraduates at MIT and are both robotics enthusiasts. a16z described the two as having both research capability and the pragmatic ROI focus of enterprise software. Put simply, they aren't just showing off technology; they're focused on whether robots can work reliably on production lines and whether the numbers add up.

There is one detail worth noting. The story Tutor tells externally is not "robots replace humans," but rather "humans first bring robots up to speed." Remote operators are part of production at this stage, not a temporary patch for a transition period. Whether capacity and model capabilities can rise at the same time, and whether this loop can keep turning, will determine whether the 1,000-robot target is a number on paper or real revenue.

A user asked a key question under the tweet: as the number of robots grows, can the human operation time required per task come down without narrowing the range of tasks the robots can handle? The announcement didn't answer this question, but it happens to be the core metric of this model.