Why this startup wants more engineers playing with robots

Steak used to be Nicole Maffeo’s favorite food. Then she started cooking several ribeyes a day as sous chef to a bossy robot. “I can’t look at another one,” Maffeo said.

The searing marathon has been taking place in a robotics workshop in the Manhattan offices of Viam, an automation software platform. There, Maffeo has been building out a small startup called Gambit Robotics that aims to make a kitchen assistant for consumers.

Viam founder and CEO Eliot Horowitz says he wants his platform to be a software layer for a new wave of robotics and automation, similar to Gambit’s project. Horowitz, who previously co-founded database giant MongoDB, cooked up the idea with Maffeo over a game of chess in Washington Square Park (Maffeo is a former member of the US Chess Team).

Horowitz thinks the world is ready for a whole slew of new robots. They won’t be powered by generative AI—the tech might serve as an interface—but he said robotics has benefited from some of the same innovations that are making generative AI possible: cheaper, more energy-efficient chips, better connectivity and batteries, more compute power.

“Right now, it’s green field,” Horowitz said. “There are way too few people who are ready and excited about building these kinds of things yet, because historically, the trope in the VC world is that hardware is hard. People are excited, but it’s early. It’s very early.”

Viam’s business is currently mostly in industrial and enterprise settings, helping businesses install and manage physical automation. Its client list is eclectic. At UBS Arena, home of the New York Islanders, its tech helps fans monitor bathroom and concession lines. It has worked with Sbarro to monitor pizza buffet longevity. It works with multiple boating companies on manufacturing and fishing insights. (“Why? Because I really like boats,” Horowitz said.) Manufacturing is a big target area for growth, he added.

In the workshop: Viam’s workshop helps engineers get accustomed to building on the platform. Located above Sesame Street on the Upper West Side (not the set of the children’s show), it’s a quirky place even by tech office standards. Two mounted robotic arms stand ready to face off over a chessboard near the entrance. Another machine steadily sands away a wooden board, apparently practicing for use on fiberglass boat parts.

Engineers have built what they call a robotic sommelier as a demo of Viam’s platform. Two mechanical arms meet mid-air to pour a wine bottle—in the case we saw, it was a bottle of Saratoga Springs water—into a tilted glass.

What makes for a good robot? Maybe you’d like a robot to load your dishwasher or do your laundry. “Everyone has something in their house they want to automate,” Horowitz said.

Whether or not a robot would be good at any of these things is not actually intuitive until you try building your own, according to Horowitz. Gripping and picking things up can be a challenge; a lack of sensors in the upper arms can make working in tight spaces difficult, Horowitz said. But robots can also operate more smoothly and fluidly than people expect, he said.

“It’s very hard to have good intuition on feasibility,” Horowitz said. “That’s where I think we can help a lot. We try to really help people understand, ‘OK, actually, you think that’s hard, but actually that’s not that hard. But the other thing that you think is easy is actually nearly impossible right now.’ I think that’s where the biggest gap is.”

Robo sous: Horowitz acknowledges that most home chefs don’t want to cede their hobby to a Jetsons-like contraption. The idea is to eventually help with keeping a pot stirring or even cut vegetables with a mounted robotic arm.

“It’s a few thousand dollars, which for a lot of people, you probably save that much money in takeout food if you’re actually cooking more at home,” Horowitz said. “A big robot that moves around in a kitchen? That’s going to be a while. That’s way more expensive.”

Right now, Horowitz and Maffeo are working toward a generative AI-powered device that mounts above a stovetop. It can upload any recipe and distill it into a stream of guidance as well as using sensors to detect temperature and imagery.

“The temperature is too low for your steak,” the voice chirped in one of these many tips. “Try turning up to medium-high heat.”

Before she started testing it on steak, Maffeo said she trialed the system on around 2,000 pancakes. “Pancakes are actually kind of hard for a visual classifier because they look white, which is a harder color for AI than red,” she said.

Horowitz hopes that startup projects like Gambit will eventually become successful enough that they will inspire more engineers to build on the Viam platform. “It’s about showing them enough examples of things working,” Horowitz said. “Getting a few companies going early now ourselves is great because it just gets more and more examples out there.”