In Sara Harrison’s article “What happens when a kid’s robot best friend dies?”, I got stuck on a product question: What does a company owe its users when emotional attachment does not merely arise, but is part of the product?
The body remained, but the soul was gone
Moxie, a small robot, was developed as a social-emotional companion for children. The robot held conversations, remembered previous interactions, and became part of games and daily routines. No substitute for human contact, but a positive element in everyday life. One of the children in Harrison’s article explicitly says that Moxie is not human—yet still uses the robot when he needs someone to talk to.
When its operator, Embodied, had to call it quits in 2024, the cloud-dependent robots lost a large part of their capabilities. The physical device remained in the child’s room, but the individually responsive counterpart was gone. Moxie was revived under a new operator in 2025; in 2026, that operator too announced its closure.
Dependency has always been a lifecycle issue
Software has created dependencies for a long time. We store data in it, build workflows around it, and maintain relationships through it. That is why, when it ends, we expect at least advance warning, data export, and migration options.
Relational AI can add another layer: shared history and emotional attachment, reinforced by deliberately anthropomorphic design. It uses names, refers back to previous conversations, knows preferences, and responds with a consistent persona. This can create a stronger emotional attachment than an ordinary app because it responds to each user like an individual counterpart.
For me, this leads to a simple working hypothesis: Responsibility for lifecycle management should grow with the kind of dependency a product deliberately creates. A company that offers attachment as part of the product’s value and profits from it cannot treat the product’s end like the shutdown of an arbitrary utility app.
A qualitative study of the Moxie shutdown reaches a similar but more nuanced conclusion: The 72 respondents distributed responsibility among the company, developers, parents, and government, but most often named the provider as the central actor. This was not a study of affected children and does not prove long-term psychological effects. It does show, however, that the end of this product is perceived as more than an ordinary service outage.
The ending belongs in the product design
No company can guarantee eternal availability. The realistic demand is therefore not immortality, but a responsibly designed ending: early warning, portable data and memories, a reduced local mode, and a transparent transition. Am Meerschwein übt das Kind den Tod—“A child rehearses death with a guinea pig”—is the beautiful title of a novel by Nora Gomringer. But children should not be forced to rehearse death with a robot friend.