The Roomba Inventor Quit After Amazon Killed His Deal. Now He's Betting Everything on a Furry Robot That Can't Talk

Colin Angle built Roombas for three decades. He watched them spread from living room to living room, becoming one of the most deployed consumer robots in history. Then Amazon walked away from a $1.7 billion acquisition of iRobot in 2024, and Angle found himself standing in a company that had just lost its future. He left. Now he is building something he believes the robotics industry has been trying to build incorrectly for twenty years.

Familiar Machines & Magic, his new company, unveiled its first product last week: a quadruped robot roughly the size of a medium dog, covered in touch-sensitive fuzzy material that borrows construction techniques from the sneaker industry and 3D knitting. It has 23 degrees of freedom, an on-device Nvidia Jetson Orin processor running a custom small multimodal model, a vision system, and a microphone array. It makes nonverbal sounds. It does not speak. There is no screen.

Angle's pitch is simple and, depending on your tolerance for technological eschatism, either clarifying or ominous: the reason companion robots keep failing is that they keep trying to talk to you.

"The next era of robotics is not just about dexterity or humanoid form," he said at the launch event, "it's about machines that can build and sustain human connection." The quote sounds like marketing until you pair it with what he told IEEE Spectrum: "I don't believe that the technology exists today for AI to talk to humans in a safe, responsible fashion." He is not hedging. He is drawing a line.

The logic is architectural. Angle has articulated what he calls the plate of glass test: if a sheet of glass between you and the device would not change the experience, the device should just be a screen. Physical presence, in his framing, is not a nice-to-have. It is the entire point. A robot that lives in your home and cannot be dismissed with a tap is a different category of object than a smart speaker, and Angle believes that difference is load-bearing for the thing he is actually trying to build: something that people feel responsible for.

This is a bet about loneliness, and Angle is not coy about it. He is building for people who want a companion but cannot or do not want a pet. He is also, whether he says it plainly or not, building for a cultural moment in which the aspiration to that kind of presence has become its own confession. Companion robots have a long history of failing to deliver on exactly this promise. Anki, which made the charming and briefly beloved Vector robot, collapsed in 2019. Mayfield Robotics' Kuri, a rolling conversational companion with personality, was discontinued the same year. Jibo, the social robot with a rotating cone head and genuinely expressive animations, died in 2018 after a funding collapse. Each of these had voices. Each had personality. Each generated genuine early affection and then was abandoned by its parent company or simply failed to sustain the relationship.

The one genuine long-term success in the companion robot space is Paro, a seal-shaped robot used in institutional care settings, particularly dementia care. Paro does not speak. It responds to touch and sound with movements and vocalizations. It has been deployed in hospitals and nursing homes for over two decades. The research on it is mixed, but the places where it works best are precisely the places where a warm body of any kind is scarce and the alternative is nothing. Angle cites this history, and the contrast is not accidental: Paro succeeded where the market of individual consumer purchases has consistently failed, because Paro was solving a specific, acute form of social absence rather than offering a general-purpose emotional gadget.

Angle's new robot is not aimed at institutions. It is aimed at the individual who wants something in their life that notices them, that they care for, that is physically present in a way a screen is not. The price, announced as comparable to dog ownership, covers an enormous range. It could mean a few hundred dollars. It could mean thousands. Angle is not saying yet, and the vagueness is probably intentional: anchoring the product in the cultural and financial category of pet ownership rather than the category of consumer electronics is a deliberate move up the ladder of perceived value.

The delivery timeline is 2027. The robotics industry has a documented history of optimistic shipping dates, and Familiar Machines & Magic is a new company with a new product. Angle has personal emotional investment in the product — he has described a D&D familiar (a conjured creature companion) as the conceptual ancestor of what he is building — and that kind of investment can cut both ways in terms of execution pressure.

The counterargument, which several reviewers have already raised, is straightforward: voice assistants are already in homes, already trusted with information and routine, and they cost a fraction of what a dog costs. A robot that cannot answer a question about the weather is strictly worse than the rectangle in your pocket for any task that requires language. The thing it offers instead — physical presence, touch responsiveness, the ambient awareness of something alive nearby — is either exactly what a certain kind of person has been missing or a very expensive replacement for a houseplant, depending on how you see it.

Angle's answer to that is the one he keeps returning to: "If this is a toy, we've failed. If this is a creature that you want in your world, then we've knocked it out of the park." It is a high bar. It also sidesteps the harder question, which is not whether the robot is good but whether the problem it is solving is one that more robots can fix. Loneliness research is robust in documenting the scope of the problem. It is considerably less settled on whether technological substitutes for human connection reduce social isolation or calcify it. Angle is betting, with a quadruped that cannot speak, that the answer is yes.

The Amazon deal that never happened is relevant here only as backdrop. iRobot, without a buyer and without a clear path to the kind of scale that would justify its existing operations, is not the company that could have built what Angle is building now. The collapse cleared the way for something smaller and stranger. Whether it also cleared the way for something real is the question that 2027 will answer.