What if the next big wave of AI isn’t about robots doing your chores but about robots that understand you?
In this episode of TechFirst, we sit down with Colin Angle, co-founder of iRobot and the creator of the Roomba, to explore his bold new venture: Familiar Machines and Magic. After putting over 50 million robots into homes, Angle is now betting on something radically different: a quadruped AI companion designed not for work, but for connection.
This isn’t a humanoid.
It’s not a vacuum.
It’s something entirely new.
- Get the deepest insights concisely on the TechFirst Substack newsletter
- Subscribe to the TechFirst YouTube channel to never miss an episode
- Or listen on Spotify or Apple Podcasts
This episode of TechFirst is sponsored by Apprentice
Did you think AI was only for digital work? Nope … AI-native manufacturing is here. This month’s sponsor is Apprentice, which offers the first AI Agent built from the ground up for agentic manufacturing. Connects to all your systems, monitors everything, automates all your processes … but keeps a human in the loop. Check it out at apprentice.io.
Watch our conversation here:
Powered by on-device multimodal AI, this “familiar” can follow you around your home, learn your routines, encourage healthier habits, and even develop a kind of relationship with you, all while keeping your data private.
We dive into:
- Why the humanoid robot race might be overhyped
- The massive untapped “emotional AI” market
- How this robot learns, adapts, and interacts like a pet
- Privacy-first AI design (no cloud streaming)
- Why form factor matters more than you think
- The future of robots in everyday life
Colin also shares why now is the perfect moment for physical AI—and how advances in reinforcement learning and edge computing are making this possible.
Transcript: Roomba CEO’s new home robot: not humanoid
Colin Angle:
One way of thinking about what Familiar Machines is, it’s kind of taking that groundbreaking price performance that allowed iRobot to build a new industry and combining it with the pioneering lifelike dynamic motion of Boston Dynamics with the animatronic characters of Disney.
John Koetsier:
Colin Angle has sold more home robots than probably anyone else alive. He co-founded iRobot in 1990—that is a while ago—and put more than 50 million Roombas into homes around the world, making pets happy all around the world.
iRobot was going to be acquired by Amazon; that collapsed. He stepped down and started building something else. Today, that something else is coming out of stealth. His new company is called Familiar Machines and Magic. His first product is a 23-degree-of-freedom quadruped.
Not a dog—a quadruped. It has a touch-sensitive coat of fur. It’s running a small multimodal model on-device. It’s not going to clean your floors. It’s not going to do your chores. It will greet you when you get home. It will follow you in the kitchen while you cook. It will nudge you off the couch if you’re sitting too long.
He’s calling it a “familiar,” and his thesis is pretty contrarian. He’s saying that the entire humanoid robotics industry—Figure, 1X, Apptronik, Tesla, Agility, and others—billions in venture capital chasing factory labor—is all fighting over a small chunk of the physical AI market.
The big chunk, he thinks, is not about physical work. It’s about emotional work. It’s about companions, connection, presence—the kind of thing that he’s saying a humanoid isn’t going to deliver.
In this episode, we get into it. We talk about what it is, how it works, and what it looks like. Before we get into that, a quick note from a sponsor. Did you think AI was only for digital work? Nope. AI-native manufacturing is here. This month’s sponsor is Apprentice, which offers the first AI agent built from the ground up for agentic manufacturing. It connects to all your systems, monitors everything, automates your processes, but keeps a human in the loop. Check it all out at apprentice.io.
So Colin, you’ve spent your entire life building robots. You’re releasing a new one right now. Tell us about it.
Colin Angle:
What we’re doing is coming out of stealth. We’ll be talking about what we’re doing and the markets we’re trying to serve. We can show some teases and hints about the robots and maybe a little bit of footage of the initial experiences.
But just to be clear, the product launch is next year.
John Koetsier:
Okay.
Colin Angle:
Not that far away, but next year. This is something I’m very excited about. I’ve been thinking about robots that have more human interaction than just floor vacuuming for a very long time. In fact, the first name of iRobot was Artificial Creatures Inc.
Now I get to actually do it.
If it’s helpful, a little bit of background: this is an amazing time to be a roboticist. Physical AI is huge. Industrial applications are big, but applications of physical AI in the home are also going to be tremendous. That means we have to figure out how robots are going to interact with people.
My new company, Familiar Machines and Magic, was set up to focus not on humanoids and their applications in industrial settings—which is about half of the five-trillion-dollar total addressable market for physical AI over the next couple of decades—but on the other half. The other $2.5 trillion is about solving challenges around interacting with people and, rather than doing physical work, doing something more like emotional work.
John Koetsier:
Interesting.
Colin Angle:
One way of thinking about what Familiar Machines is, it’s taking that groundbreaking price performance that allowed iRobot to build a new industry and combining it with the lifelike motion of Boston Dynamics and the animatronic characters of Disney.
That’s what a familiar machine is. And the “magic” part is the generative AI that brings these things to life.
John Koetsier:
That’s a very big promise. That’s huge. You just named three really critical components. Boston Dynamics is incredible, Disney does incredible work, and you add another component as well.
Wow.
Colin Angle:
Yes, and it’s needed. Bringing physical AI into the home is exciting, but the form that first solution takes may not be what people expect.
In a lights-out factory, it doesn’t matter what you look like. Your form is driven by efficiency. As soon as you bring people into the equation, it gets complicated.
If I have a smart speaker that looks like a puck, I expect to talk to it and hear it play songs. I don’t expect it to move across my floor and pick up books.
But if I make that same thing look humanoid, all bets are off. Can it talk to me? Does it have emotions? Can it clean my room? Expectations explode.
So you have to be very intentional about the expectations you create. You also have to be very careful about privacy and security—just like with Roomba, but even more so.
We also have to think about engagement. What will these robots do to become part of people’s routines?
If you like, I can play a short video to give you an idea.
John Koetsier:
Please do.
Colin Angle:
All right, here we go.
Video Narrator:
Place in your life, in your routines, in the small moments that make up a day, learning and anticipating when you want support or simply company. Over years, the relationship deepens.
You care for it, and it cares for you.
It learns when to coach.
John Koetsier:
Wow.
Colin Angle:
Right? We can make this real. The idea of a robot that greets you, nudges you toward healthier routines—I’ve been studying the home for decades.
Any robot is a financial investment. There was an opportunity with vacuum cleaners, less interest in mopping robots or folding clothes. But people spend a lot trying to find joy and companionship.
What we’re building isn’t really a pet, but it’s pet-like. It can learn your schedule. It can notice if you’ve been doomscrolling and suggest a walk.
At its core, it’s not trying to replace human relationships. It’s trying to get you out into the world to connect with others. If it builds a relationship with you, it has more influence to encourage healthy behavior.
It can follow you into the kitchen, keep you company, watch sports with you, even have opinions.
John Koetsier:
That’s really interesting. You mentioned Roomba-scale cost—not $5,000 or $10,000. You’re not releasing pricing yet, but maybe $1,000 to $2,500.
But you’re describing something physically capable—walking, uneven terrain. That’s not easy. Boston Dynamics’ Spot does it, but that’s expensive.
Colin Angle:
We’ve said the cost of ownership of a familiar is similar to owning a pet. If you can afford a pet, you can afford a familiar.
Interestingly, many enthusiastic early adopters are pet owners. They like caring for things. There’s a bidirectional relationship—it cares for you, and you care for it.
John Koetsier:
I’m fascinated by your thought process. If something looks like a puck, expectations are limited. If it looks human, expectations skyrocket.
Colin Angle:
Exactly. It’s very hard to manage those expectations. There may be a future for humanoids in the home, but that’s a second step.
What we’re doing is easier, more acceptable, and laser-focused on a real human need. When we get to humanoids, we want them to feel familiar, not uncanny.
This is a platform. You don’t get credibility until people build something amazing on it. Our journey starts with familiars. We build trust and understand what’s acceptable.
To address privacy concerns, all compute is done onboard. We don’t stream video or audio to the cloud. We’re not even fully comfortable with conversational AI yet. Instead, familiars communicate in pet-like ways—sounds, motion, expression.
That alone can create powerful interactions and removes major privacy concerns.
If users choose to share data, it’s optional, transparent, and for their benefit. They can opt out anytime, and we delete their data.
With these safeguards, plus valuable functionality at a pet-level cost, it starts to feel much more reasonable.
John Koetsier:
Very cool. It can hear, see, and probably has multiple models running onboard. What other senses does it have?
Colin Angle:
We use array microphones with speech-to-text, stereo vision, and rangefinders feeding a multimodal model trained with a personality.
It also has an emotional state and memory formation. That feeds a high-level goal system, updated about once per second, which drives a fast behavior model for movement and interaction.
With 23 degrees of freedom and powerful processing, there’s a lot packed into this robot. It’s the classic iRobot challenge—putting 500 pounds of capability into a five-pound bag.
John Koetsier:
It can understand and learn from interaction, like a pet.
Colin Angle:
Yes. It learns what behaviors get positive reactions and adapts. We call the user the “guardian” rather than owner.
If something doesn’t make the guardian happy, it adjusts. It reads context—like whether you’re busy or having a bad day—and responds appropriately.
It won’t always behave the same way. It’s well-behaved, but it’s also dynamic.
John Koetsier:
Super cool. You had all these new tools—AI, actuators, models—and chose a quadruped. Why?
Colin Angle:
We didn’t want to be a dog or a cat because of expectations. We chose an “abstract bear.”
That allows for discovery and surprise. The guardian and familiar explore what roles it can play together.
There are many opportunities—helping kids, encouraging activity, offering alternatives to screen time.
John Koetsier:
That ties into your platform idea. If you nail the first version, there’s room to expand—cleanup, elder care, medication reminders.
Colin Angle:
Exactly. Elder care is huge—activity monitoring, medication reminders, improving quality of life.
Without changing form, we can support many applications. The underlying AI and control systems are flexible.
Reinforcement learning has changed everything. Instead of complex manual modeling, we train behaviors—and they work.
It’s an incredible time to build robots.
The product launches next year—May the 4th. A little Jedi magic.