When Anyware Robotics began commercializing its Pixmo industrial robot in 2025, CEO Thomas Tang was not looking for a customer willing to buy a finished product. He was looking for one prepared to help build it.
That customer was Saddle Creek Logistics, a major US third-party logistics provider. The company was looking for a way to automate unloading floor-loaded shipping containers, a physically demanding warehouse task.
Saddle Creek stood out because it wasn't new to robotics. The company already had multiple robotic systems in operation, giving its innovation and operations teams a realistic understanding of what deploying an early-stage system would involve.
"We both wanted to create a more sustainable working environment," he said. "They were very supportive from the very first meeting."
That understanding mattered because Tang sees the gap between the lab and the customer's real-world environment as one of the biggest challenges in developing physical AI. Software AI developers can use the products they are building themselves, Tang said, but Anyware's engineers don't run the industrial facilities they are designing robots to work in.
Tang said the entire Anyware team spent months at or near the facility during the early deployment, observing how Pixmo behaved and how well it integrated into Saddle Creek's workflow. The team rented accommodation nearby and, when there weren't enough rooms, they slept on the floor.
The first deployment took several months to stabilize. Pixmo depended on interconnected hardware, software and data-pipeline components, and a problem with any one of them could stop the machine, Tang said. The team worked through bottlenecks component by component and added backup systems around weak points to improve reliability.
Those lessons could only come from real-world deployment, Tang argued. "You will never get it from a textbook," he said. "There's no shortcut."
Keeping Saddle Creek on board through those early problems required transparency about the state of the technology: Anyware was open about what was going wrong, how long fixes might take, and the work it was doing to resolve them, Tang said.
"We didn't want to give them unreasonable expectations," Tang said. "We let them know this was just the beginning, and they were really supportive."
The deployment process has since accelerated as Pixmo has matured. What initially took several months fell first to around a month, then two weeks and, in some cases, days. Saddle Creek has expanded Pixmo beyond the initial facility to multiple sites.
For Tang, that expansion is the strongest validation of the first year of commercialization: the customer that helped Anyware work through Pixmo's early problems was willing to deploy it again.
Read the full article on why general-purpose robots start with the jobs nobody wants here.
