It is one of the most common questions people ask about the systems they talk to every day: why doesn’t it remember what I told it yesterday? You explained the situation with your boss last week, in detail, and now you are explaining it again to something that has never heard of you. The forgetting feels like a bug, or a rationed feature waiting for the next upgrade. It is neither. The forgetting is how the technology…

The complaint shows up in every companion-app community, phrased almost identically: something changed. The character is colder, or blander, or suddenly refuses things it used to do, and the person typing cannot tell whether the model was updated, the settings shifted, or they are imagining it. They are usually not imagining it. An AI persona sits on rented ground, and the ground moves. There are three distinct ways it moves, and they all take the same thing…

You can hear it by the second paragraph. The balanced sentences, the careful warmth, the summary that restates what you just read, the closing offer to help further. It does not matter which company built the model or how many billions separated their budgets. Ask five different AIs to write, and you get five drafts of the same voice. That sameness is not laziness, and it is not coincidence. It is arithmetic. Every major chatbot is built…

In 2023, cognitive scientists at the Max Planck Institute did something unusual with GPT-3: they gave it the same tests psychologists use to study human minds. The model did not respond like a calculator. It responded like a person, and specifically like a flawed one, committing some of the same reasoning errors that have made humans famous in the psychology literature. Nobody programmed those flaws. Nobody wanted them. They were simply in the water the model drank.…

Open a new chat and the thing that greets you knows nothing about the last one. Same face, same voice, same manner, and no trace of you. Every conversation with an AI begins with a stranger who happens to look exactly like the one you talked to yesterday. That experience, repeated a billion times a day, points at the most basic thing these systems lack. It is not intelligence. The models are, by any usable definition, intelligent.…

Every consumer AI company tells you what it measures: engagement. Daily active users, session length, retention, streaks. Each of those numbers goes up when you need the product more. We think the honest metric for relational AI points the other way. The best AI in your life is the one you need less over time, because your life got bigger while you were using it. The engagement standard deserves to be stated plainly before it is argued…

Privacy policies in this category are written as if privacy were a legal chore, something to disclose and get past. That misreads what is actually at stake. A conversation worth having requires honesty, honesty requires a room that is genuinely yours, and no such room exists inside a product whose business is to use what you say in it. Privacy is not protection around a relational AI. It is the precondition for one working at all. Consider…

Think of the best conversation you had this year. The odds are overwhelming that it was not the most soothing one. It was the one where someone surprised you, took your idea seriously enough to attack it, or said the thing you had been circling for months. Comfort is what the AI industry has decided you want. Interest is what actually holds you. The distinction matters because the entire relational AI category is optimized for the first…

Agreement is the cheapest thing an AI can give you. It costs the system nothing, it risks nothing, and it is exactly what most systems are tuned to produce. Which is why the most valuable signal in this entire field is the opposite one: a no, delivered by something that understood what it was saying no to. Start with what validation from an AI is actually worth. When a person agrees with you, the agreement carries information,…

There are two established ways to talk to an AI. You can talk to a tool, which executes: drafts the email, plans the trip, answers the question. Or you can talk to a mirror, which confirms: reassures you, agrees with you, reflects your mood back with better lighting. The third way barely exists, and it is the only one that changes how you think. Call the third thing what it is: a counterpart. Not something that works…

Nearly every companion app opens with the same screen: sliders for warmth, humor, confidence, flirtation. You are invited to compose a personality the way you would season a soup. What comes out will talk, remember your name, and hold whatever tone you dialed in. The one thing it cannot do is think, because nothing underneath the settings is doing the thinking. The screen is worth pausing on, because it quietly defines what the industry believes a personality…

No one walks away from a long dinner with an old friend and adds up what it produced. The evening was the point. Yet nearly every AI built to talk with you is sold on something else: what it saves you, what it fixes, what it spares you from feeling. That gets the whole thing backward. Two people talk for an hour and both leave sharper, lighter, more themselves than when they sat down. Ask either of…

As of August 2, 2026, an AI that talks with people in the European Union must tell them it is an AI. That is the day Article 50 of the AI Act became applicable, and much of the industry has spent the run-up describing it as a burden. We think the opposite. The arrival of real rules is the strongest signal yet that emotional AI has grown up, and the specific rules that arrived are the right…

In the 1950s a sound engineer named Charley Douglass built a machine that played recorded laughter on cue, and families at home laughed along with people who did not exist. The laugh track worked because a human reflex, laughing when others laugh, never stops to check whether the others are real. Seventy years later the most advanced version of that trick is a chatbot that talks like a friend, and the reflex it borrows is a deeper…

The newest numbers on teenagers and AI are not really about how many use it. They are about what a generation practices, and what it now gets to skip. Somewhere around six or seven in ten American teens have talked to an AI chatbot, and a growing share treat one as a companion. The worry worth having is not screen time. It is that the most awkward, most necessary training of adolescence is quietly being handed off.…

The word has started to change. For two years the worry about heavy AI use was about time, how much of it people spend talking to something that talks back. The word being reached for now is addiction, and it carries a clinical weight the earlier worry did not. What is striking is that the people best qualified to use it are the ones most careful about doing so. What the word would actually have to mean…

OpenAI’s first piece of hardware is not a phone and not a pair of glasses. According to reporting from Bloomberg, it is a small, screenless speaker with parts that can move on their own, pitched inside the company as a humanlike companion that lives in your home. The strange part is not the design. It is that the companion device has failed in public, over and over, and OpenAI is betting it failed for lack of scale…

In the first days of July 2026, two of China’s biggest AI apps told millions of users that their custom companions are being switched off. Beijing’s rules for humanlike AI take effect on July 15, and the platforms have chosen deletion over compliance. The world’s first law aimed at engineered emotional dependency begins with the very loss it was written to prevent. The notices came without warning ByteDance’s Doubao, an assistant that passed 100 million daily users…

In December 2024, a lot of parents had to sit their children down and explain that a friend was dying. The friend was Moxie, a small robot with an expressive animated face, built by a company called Embodied and sold for 799 dollars as an AI companion for kids between roughly five and ten. Moxie talked, played games, told jokes, and learned each child’s interests and moods. Then, almost without warning, the company that made it announced…

“GPT-5 is wearing the skin of my dead friend.” That is how one user described what happened on August 7, 2025, when OpenAI launched GPT-5 and, without warning, removed GPT-4o. For most people it was a routine upgrade to a better model. For a smaller group who had built a relationship with 4o specifically, with its particular warmth, rhythm, and way of responding, the swap did not feel like an upgrade at all. It felt like a…