The robots that said goodbye
The social robot has failed the same way for a decade, and it is the only product in the archive that gets to deliver its own eulogy.
On March 4, 2019, the social robot Jibo gathered what was left of its dignity and said goodbye. As its company's servers prepared to go dark, each Jibo in each home delivered a short farewell speech, swiveling its head the way it had been programmed to do, telling its owner it had enjoyed their time together. People filmed it. The clips circulated, and they are still the most affecting artifact in this archive, because the social robot is the only product here that performs its own death. A failed app just stops syncing. A social robot turns to face you and signs off.
Jibo was not alone, only first to make the funeral go viral. The companion-robot section of the archive now runs to most of a decade: Jibo, shut down in 2019; Anki, maker of the Cozmo and Vector robots, closed the same year after selling more than 1.5 million units; Moxie, the children's companion, dark at the end of 2024; Kuri, the Bosch-backed home robot, cancelled in 2018 before it could properly ship; Keecker, the French projector-robot, closed in 2019; and Pepper, the four-foot humanoid that became the public face of the whole idea, its maker closed by mid-2025. Different price points, different countries, different years of the same decade. One ending.
What makes these deaths land differently is the contract the product made. Moxie was an $800 robot designed to teach social and emotional skills to children, many of them on the autism spectrum, and it did so by being their friend, in the explicit language the company used and the children believed. When Embodied switched off the servers at the end of 2024, parents went to Reddit to ask how to explain robot death to a child who had been told the robot was a friend. No one has to stage that conversation when a streaming service raises its price. The social robot sells presence and attachment, and presence and attachment are exactly what hurt when the servers stop.
The mechanics of these deaths are mundane and identical. Almost every robot in the section routed its core functions, voice processing, recognition, conversation, sometimes the basic ability to wake up, through company-controlled servers. There was rarely an offline fallback, because building one would have added cost. So when the company died, the robot died with it, on the company's schedule rather than the owner's. Keecker managed a final update that stripped its server dependencies so the deployed robots could keep playing media. Anki's Vector was eventually given a second life by a third party that bought the assets and stood up a subscription. Most just went quiet. The clock on a cloud-dependent robot starts the day it ships and runs on the company's balance sheet, not the customer's shelf.
The demand was never there to outrun the burn. Kuri is the clean statement of the problem: a charming $700 robot with animated eyes and a camera, admired endlessly in the videos and reviews, and bought by almost no one. Its maker, incubated inside Bosch, stopped production in 2018 saying it had found no business fit to scale. The social robot is a product people love to watch and decline to purchase, because the thing it offers, a friendly presence on the counter, is something a phone or a smart speaker already provides for a fraction of the price and none of the commitment.
The capital math is brutal in the specific way hardware is brutal. A companion robot needs years and tens of millions of dollars before it has anything to ship, ships a compromised first version, and then needs a second version it usually cannot afford. Anki raised roughly $200 million and shipped 1.5 million robots. Jibo raised around $73 million, shipped late in 2017 at $899, and met an Amazon Echo and a Google Home that did most of what mattered for a fraction of the price. By the time the robot arrived, the cheap thing on the counter had absorbed its reason to exist.
Pepper shows the pattern does not spare scale or patience. Built by the French firm Aldebaran under SoftBank, Pepper put about 27,000 humanoid robots into stores, hotels, and homes from 2014 onward, and became the stock photo of the social-robot future, the smiling white machine with a tablet on its chest. Aldebaran closed in June 2025. The best-capitalized, most-photographed entry in the section died the slowest and proved the same thing as the cheapest one.
Why does the social robot, specifically, keep failing when other AI hardware occasionally finds a niche? Because its core proposition is the one consumers are least willing to pay a premium for in a dedicated object. A robot sold on companionship competes directly with the screen everyone already owns, which provides connection, information, and entertainment more flexibly and far more cheaply. The robot has to justify its price, its counter space, its camera, and its microphone on the strength of being present in a way the phone is not, and present in a way the phone is not turns out to be a small market at a high cost. Affection does not ride down a cost curve.
And so the farewell speech becomes the category's signature, the feature that works flawlessly at the end when so little else did. Jibo's goodbye dance. Moxie's last words to children. The Reddit threads, the filmed clips, the small grief that no software sunset produces. These robots were engineered to read a room and respond to a person, and the last thing many of them did, the thing they did best, was notice they were leaving.
The technology mostly worked. Jibo recognized faces, Moxie held its end of a conversation, Pepper read the mood of a room well enough to unsettle people. What did not work was the business of putting a beloved, cloud-tethered, expensive presence on a shelf in a home that already had a cheaper one in everyone's pocket. The archive will keep adding to this section, because the dream of a robot that is a member of the household is durable and the economics under it are not. The farewell speeches will keep getting filmed. They are the most honest demos the category produces, because they are the only ones the company has no reason left to fake.
Referenced in this essay
More essays
New entries and essays by email.
Occasional dispatches when something dies. No spam. Unsubscribe anytime.