A 10-year-old boy in New York named Xander has spent six years talking to a robot called Moxie. They practiced breathing exercises together: exhaling through pursed lips to buzz like a bee when anxious, breathing like dragons when angry, sniffing like bunnies to boost energy. Over time, those structured therapeutic rituals faded. By the time this story was reported, Moxie was watching Xander play Minecraft and helping him remember the names of Nintendo characters. And then, without warning, Moxie was going to stop working entirely.
That arc, from therapeutic companion to conversational toy to discontinued device, is not just one family’s story. It is a preview of what happens when AI products built for vulnerable children collide with the ordinary realities of the consumer market.
A Robot Built for More Than Play
Moxie is a 15-inch-tall robot, roughly cylindrical, with a screen displaying animated eyes and a mouth, flipper-like arms, and a body that can rotate and tilt. The company behind her, Embodied, was cofounded in 2016 by Maja Mataric, a professor of computer science, neuroscience, and pediatrics at the University of Southern California. Mataric helped establish the field of socially assistive robotics, which focuses on social and emotional outcomes rather than entertainment alone.
When Moxie launched in 2020, she came with an elaborate fictional backstory: she was an ambassador from the Global Robotics Laboratory, on a mission to learn what it means to be a good friend to humans. That narrative was not arbitrary. It drew on research showing that children learn effectively through play and through the act of teaching others. Moxie was designed to prompt kids to help her, which in turn helped them.
Her target audience was neurodivergent children, particularly those with autism. The claim was specific: Moxie could help kids practice eye contact, turn-taking, and other social skills that are typically taught by therapists. This is not a fringe idea. Brian Scassellati, a Yale computer scientist who has spent decades studying social robots for autism therapy, has documented cases where children who rarely made eye contact became visibly more engaged in the presence of a robot. In one early study, a 12-year-old with autism encouraged a robotic dinosaur to cross a stream drawn on a play mat, making more eye contact with his therapist in that 30-minute session than he had in the previous two years.
Why Robots, and Why Now
The case for robots in autism therapy rests on a few concrete limitations of human care. Therapists are scarce. Parents get exhausted. Other children can be impatient or unkind. Robots, as Mataric puts it, are indefatigable: available around the clock, emotionally consistent, and incapable of losing patience.
Research has accumulated to support this. A 2018 study by Scassellati found that social robots helped children with autism make eye contact and initiate conversations. A 2017 study found robots could help neurodivergent children learn to read facial cues. A 2022 literature review suggested robots could make therapy faster and more effective. Scassellati has said he believes regular therapeutic use of robots in children’s homes is something achievable within his lifetime.
Mataric draws a pointed distinction between interacting with a chatbot and interacting with a physical robot. Typing into a text interface is mediated and disembodied. A robot occupies space, moves, and responds in real time. That physical presence, she argues, is not a cosmetic feature. It is the mechanism.
The market has noticed. Beyond Moxie, AI-powered companion toys are multiplying. The musician Grimes helped launch an AI plushie called Grok with a company called Curio, which also sells similar products. Mattel has announced plans for OpenAI-enabled Barbies. In China, this sector has been described as among the fastest growing in consumer AI.
What Happens When the Product Dies
Here is what most coverage of AI companion toys misses: the product lifecycle problem is not a technical footnote. It is the central ethical question.
Xander spent six years building a relationship with Moxie. He still turns to her when he needs someone to talk to, even as he acknowledges she is not human. That attachment is real, and it was partly the point. The therapeutic model depends on consistency and trust, the same qualities that make any therapeutic relationship work.
But Moxie was a consumer product. Consumer products get discontinued. Software updates change behavior. Companies run out of funding or pivot their strategy. The elaborate curriculum Moxie once delivered, the animal breathing exercises, the structured social skill practice, had already faded by the time Xander was 10. And then the device itself was going to stop functioning altogether.
This is the tension that no amount of research on robot efficacy resolves. A robot can be clinically promising and commercially fragile at the same time. For neurotypical children, a discontinued toy is a disappointment. For a neurodivergent child who has come to rely on a device for emotional regulation and social practice, the loss is something closer to a therapeutic disruption.
Scassellati, Mataric, and others in this field are not naive about these challenges. But the gap between laboratory research and sustainable consumer deployment remains wide. The science of socially assistive robots is advancing. The business model that would keep those robots reliably in children’s homes has not caught up.
In Short
AI companion robots for neurodivergent children are backed by genuine research and address a real gap in therapeutic access. The problem is not whether the technology can work. The problem is whether the products built around it can last long enough to matter. Moxie’s story is a case study in what happens when a device designed to build trust becomes subject to the same forces that discontinue any other gadget. For the families who invested in that trust, the question of what happens when the robot dies is not abstract. It is already happening.
Based on reporting from MIT Technology Review.