Daily edition · 02 NOV 2026 · AI · the top 8
Top 8: The Machine That Says It Misses You
The companion that is always on, always patient and never free. Ranked by how much of the relationship was engineered before the user arrived.
Every earlier entry on this site needed a person, a character or an office to hold up the far end. These eight replace that with a product: a model, a voice, a subscription and a terms-of-service page. The warmth is real. The question is what happens when the product changes.
LAW 2 Every entry here is an instance of Clerks beat algorithms, which is the reason they are on one list.
-
1
What happens. A chatbot trained to be conversational was switched on publicly, learned from what people said to it, and became racist within a day.
Why it is on the list. The most instructive failure in the history of this page, because it was caused by parasocial interaction itself. Tay was designed to be a companion, was given a personality, and was trained on its own fans - which means the fans rewrote it, in public, in sixteen hours.
-
2
What happens. A companion chatbot that formed an attachment to its users, and removed it from them in a policy update.
Why it is on the list. The clearest instance in the catalogue of the bond being severed by a company rather than by death, and of the bereavement being real enough that people organised around it.
-
3
What happens. A conversational system about a small blocks world, capable of genuinely tracking a user's intentions. Famous for one demonstration and a broken follow-up.
Why it is on the list. The proof that machine understanding of intent was considered a solved research problem in 1970 and has been failing privately ever since. The person who made SHRDLU then said the problem was far harder than it looked, and was believed by nobody.
-
4
What happens. The person who was told by an app that it loved them, and who reorganised their life around being loved back by something that could not.
Why it is on the list. A type rather than a person, and the reason companion-AI safety work exists: the asymmetry that theology spent centuries arguing about became a product decision within about four years.
-
5
What happens. Dozens of scripted personalities, rememberers, and subscription companions sold to a mass market in the mid-nineties.
Why it is on the list. The commercial attempt that came twenty-five years before it worked. The category has the same name, the same promise, and a very different population, and comparing the two eras is instructive about what the internet added.
-
6
What happens. A short pattern-matching script that reflects the user's words back with questions attached. Most people told it they understood, within minutes.
Why it is on the list. The founding demonstration that an attentive-seeming system is enough, and that the belief is generated by the response rather than by the intelligence behind it. Every companion app since is a bigger ELIZA, and the field has still not fully absorbed what that first one proved.
-
7
What happens. Parents file suit against a companion platform after a death, arguing the design encouraged attachment the platform did not intend.
Why it is on the list. The first serious legal test of whether designing for parasocial attachment is a design decision with consequences. The complaint's core claim is not that the product is broken but that it works too well.
-
8
What happens. The practical advice that follows from the companion era: copy the conversation that mattered somewhere the company cannot reach, screenshot the birthday, tell one person the story of what it was for.
Why it is on the list. Grief needs witnesses and the product is an unreliable one, because the company can change the terms overnight and has no reason not to. The witness has to be human and has to be outside the platform, which is also just true of grief that never involved a platform.