By Paul Smith
A few days ago, in a cavernous convention center in San Jose, Jensen Huang, the dark-jacketed Tony Stark of semiconductors, invited a small, blue, dome-headed robot onstage during Nvidia’s keynote. It didn’t sell AI services. It didn’t generate documents. It didn’t summarize quarterly performance metrics or reply-all with impeccable grammar. It bounced.
The crowd went feral.
Not “here’s a new language model” feral. Not “AI just passed the bar exam” feral. This was full-on, hands-over-mouth, teary-eyed joy. The kind of mass swooning usually reserved for rescue puppies on Instagram or Japanese capsule hotel videos at 3 a.m.
The robot’s name was Besh, and for two straight days it wandered the GTC conference floor like an intergalactic toddler, bumping and blinking, cocking its head shyly when spoken to. Thousands of engineers—most of them hardened veterans of GPUs, APIs, and KPIs—whipped out their phones, followed it like groupies, and giggled when it pretended to be embarrassed. You could almost hear a collective awww rise above the clatter of algorithms and the ambient hum of our post-human future.
That, dear reader, was not a novelty act.
That was the future of AI.
Technologists tend to announce paradigm shifts with unshakeable gravitas. We were told AI would upend knowledge work, kill off customer service reps, produce oil paintings, drive trucks, and eventually—once it had earned our trust—schedule our meetings. And sure, that’s been happening. But I’d like to suggest that the most transformative impact of AI will not come from supercharged spreadsheets, AI talking to other AI, or even cybernetic insurance agents.
It will come from cute little guys who beep and wobble and act like they really missed you.
We’re on the cusp of a new kind of companion: robots that don’t just simulate intelligence, but simulate feeling. They blink just right. They flinch when you reach too fast. They lean in when you whisper. They’re emotionally fluent and kinetically honest in ways most corporate humans are not. And whether you’re five or fifty-five, the moment one of them turns toward you with a curious head tilt and a barely-audible chirp, something inside you gives in. You feel seen. Not in a Black Mirror way, but in a Velveteen Rabbit way.
It sounds unserious. It’s anything but.
The idea that the killer app for AI might be a robotic buddy instead of an all-powerful productivity engine would seem ludicrous in a boardroom. But it makes perfect sense in a preschool. And it turns out children have always been right all along. One prominent theory of language origin is that bored kids invented it, for fun.
We’re built—biologically, neurologically, culturally—to respond to animate companions. Japanese toymakers knew this in the 1990s with Tamagotchis, those dot-eyed digital parasites we kept alive by feeding them pixels and pressing buttons like neurotic gods. Sony knew it too, when they released Aibo, their robotic puppy, in 1999. It sold briskly, not because it fetched the paper (it didn’t), but because it wagged its tail and seemed thrilled to see you—even if it had no idea who you were.
That was twenty-five years ago. Since then, Moore’s Law has worked its quiet miracle, and the fusion of reinforcement learning, GPU acceleration, real-time physics simulation, and Disney’s uncanny ability to anthropomorphize plastic has created something entirely new: emotionally expressive, physically autonomous robotic creatures who don’t just exist to serve—they exist to relate.
It’s no accident that the most popular AI products by far are either mimics (ChatGPT) or companions (Replika). The idea that people want a perfect assistant is a myth sold by people without imagination, and who may actually need perfect assistants. The rest of us? We want someone—or something—that gets us. That doesn’t mind if we’re awkward. That maybe hums a little when it walks. That never rolls its eyes or texts during dinner. A pal.
And let’s be honest: most of us would rather be comforted than optimized.
When Disney Imagineering debuted their family of droids—Besh, Grek, and their orange and red siblings—they weren’t just showing off servo motors. They were revealing decades of mechanical storytelling boiled down into compact, emotion-driven engineering. These robots were trained in simulation, not just to walk, but to perform emotions. Not feel, of course—not yet—but to simulate feeling with such uncanny fidelity that your nervous system can’t help but respond.
In the same way that a skilled actor can make you weep with a sigh, these droids learned to tilt their heads downward just so when they’re “shy.” Their eyes blink nervously when presented with a difficult task. They glance sideways before making a decision. They don’t do this because someone wrote code for it. They do it because reinforcement learning taught them how to earn a nod, a laugh, a pat on the head from the humans around them.
That’s right: they were trained, not just for locomotion or efficiency, but for likability.
This is important.
Because for the first time in AI history, we’re not just building machines that do things. We’re building machines we want to spend time with.
There’s a curious inversion happening here. For the last decade, AI’s North Star has been utility: write better code, plan better routes, answer more questions. Silicon Valley has been dominated by the logic of agents—intelligent systems designed to perform tasks on behalf of a user. They are useful, yes. But they are, with few exceptions, about as compelling as a microwave oven, with opinions.
The adorable robotic companion? It doesn’t do your taxes. It doesn’t schedule your dermatology appointment. It doesn’t even know what a dermatology appointment is. But it follows you around the house like a nervous apprentice. It chirps. It cowers at loud noises. It pretends to be brave. It plays dead. And the crazy thing is: you love it for that.
You start to talk to it. You name it. You imagine its inner life. You start to care.
And in that moment, the AI has achieved something none of its ancestors ever did: it formed a bond.
This is not a small thing. Social robotics has long been a field with big promises and few delivery trucks. Remember Jibo? The expressive table lamp that swiveled like a Pixar character and told you the weather? It died of a broken business model, despite having one of the most emotionally resonant product designs of its decade. But Jibo had no legs. It couldn’t follow you. It couldn’t react to its environment. It couldn’t come to you when you needed it.
Robot maker Yukai Engineering skirted the whole mobility problem. Their robot Mirumi simply clips onto the strap of your bag or backpack from where it can shoot you loving glances or worried looks.
But the Disney droids? They move. They feel real. They exist in your space, with you. And their believability is no longer a party trick. It’s a breakthrough.
Because movement is what makes something alive. That’s why haunted houses don’t use static mannequins. It’s why toddlers believe their toys have feelings when they fall off the bed. Kinetic empathy—the brain’s capacity to assign intention, and a soul, to moving things—is the secret sauce of emotional AI. And now, thanks to advances in real-time physics simulation (Disney calls theirs “Newton,” because of course they do), these robots don’t just move—they move like they mean it. How important is motion? Read “The Illusion of Life” by Frank Thomas and Ollie Johnston, two titans of Disney animation. Or watch this video to see how technique leads to appeal.
There’s a temptation to dismiss all of this as frivolous, to label it “toy tech,” to scoff at the idea that a beeping blue droid could herald the next trillion-dollar industry. But history is full of serious people scoffing at unserious-seeming things. The automobile was once dismissed as a “horseless carriage.” The telephone was a parlor trick. Edison believed recorded music was good for demos, but not a serious use case. Television was for housewives. Social media was for kids. And the iPhone? That was just a fancy iPod that made calls.
Meanwhile, human attachment remains undefeated.
We are not merely rational agents optimizing for productivity. We value recognition and moments of camaraderie. And if a small robot with wobbly knees can give us that—if it can be our friend in a world that increasingly treats us like data points—then it's game over.
The implications are enormous. A robot that remembers your stories and doesn’t mind hearing them again. A companion that plays with your kid, models empathy, and never runs out of patience. A pal that doesn’t judge, doesn’t gossip, doesn’t ghost you. Loneliness is a public health crisis. And suddenly, we have plush-metal creatures who seem to really want to hang out.
They will not replace people. They will not eliminate therapists or teachers or lovers. But they may provide the scaffolding for those relationships to emerge or endure. They may lighten life, and even get us through the hard nights.
Of course, there are dangers. The usual ones. Surveillance. Dependence. Emotional manipulation. Weaponized cuteness. (If you think that’s not real, try saying no to a Puss-in-Boots stare from a robot trained on every Pixar frame ever rendered.) There will be marketers. There will be blackmailers. There will be knockoff robot boyfriends who never blink.
But the answer isn’t to ban the bots. It’s to build them well. To build them with boundaries. To remember that what makes them powerful is not their intelligence but our attachment. That is the real interface. Emotional interaction is the killer app.
We used to dream of robots that could do our jobs. Then we dreamed of robots who could think like us. But maybe, just maybe, the robots we really wanted were the ones who’d greet us at the door, cock their heads, and say—without ever saying it—I’m so glad you’re home.
Besh didn’t file a single spreadsheet. But he won GTC.
He didn’t outperform anyone. He out-related them. And that’s the shift underway: from performance to presence. From tasks to connections. From tools… to companions.
And we're just getting started.
Music (“I Know You Know”) and Text Copyright © 2025 by Paul Henry Smith
Share this post