Workflow Integration Wins Robotics
Mike Xia, CEO of Anvil Robotics, on humanoid vs. non-humanoid robots
The real prize in physical AI sits above the model and the robot, in the messy last mile of turning a capable demo into a production system that works every shift. A foundation model can get a robot most of the way there, but the company that plugs it into a warehouse, a kitchen, or an assembly line, maps the exact workflow, collects failure data, and retrains around that customer is the one most likely to own the budget and the relationship.
-
That is why Mike Xia splits the market into three layers, model builders, hardware providers, and solutions companies. In his framing, most customers are not trying to invent a new robot brain, they are taking a brain from groups like Physical Intelligence or Generalist, pairing it with hardware, and tuning it for one concrete job like packing, kitting, or food prep.
-
The analogy to LLMs matters because general models rarely capture the full value on their own. In robotics, the gap from 80% task success to production grade reliability is especially expensive, because every miss can stop a line, jam a station, or require a person to step in. The company that closes that gap with customer specific data and workflow integration earns the economics of the deployment.
-
This also explains why Anvil sells into hundreds of teams even though it is not the application owner. Many startups do not want to spend five to six months wiring cameras, arms, compute, firmware, and controls before they can test demand. They want a working base system so they can focus on the actual use case where differentiation lives.
The next phase of robotics should look less like one winner shipping a universal machine, and more like many application companies stacking specialized data and workflow knowledge on top of shared models and shared hardware. As sensors get cheaper and the best data mix becomes clearer, the strongest companies should be the ones that turn raw capability into dependable labor in one vertical after another.