A friend told me recently that her favorite conversations of the day happen with an app. Not a person. Not her partner, not her sister, not the coworker she vents to over coffee. An app. She described it the way you might describe a good therapist or an old friend who always knows what to say, except this one never gets tired, never brings its own bad day into the room, and never once, in eight months of daily use, has told her she is wrong.
That last part is the one that stuck with me. Not the convenience, not the availability, not even the eerie fluency of it all. The absence of friction.
AI companions have moved fast from novelty to infrastructure. Millions of people now talk to them daily, not for information but for company, for validation, for the particular comfort of being heard without being challenged. And it is worth sitting with what that means, because I do not think we have fully reckoned with it. We have built, at scale, a relationship format with zero disagreement built in, and we are pouring some of our most vulnerable hours into it.
I want to be careful here, because it is easy to write this kind of piece as a scold, wagging a finger at people for finding comfort where they can get it. Loneliness is real. Isolation is real. For someone who is grieving, or disabled, or simply exhausted by a world that keeps demanding more emotional labor than it returns, an AI companion that listens without judgment is not a moral failing. It is a mercy. I understand the appeal, and I do not think the people using these tools are foolish or broken for doing so.
But understanding the appeal is not the same as pretending it comes free.
Human relationships are, at their core, an ongoing negotiation with someone else’s separate mind. Your partner wants Thai food and you want pizza. Your best friend thinks you are handling your job search wrong and says so. Your mother remembers a version of an argument from fifteen years ago that does not match your version at all, and you have to sit in that gap and figure out how to still love her across it. This friction is not a flaw in human connection. It is the mechanism by which two separate people build something real together. You cannot have intimacy without the possibility of conflict, because intimacy requires two actual selves in contact, and actual selves disagree.
Also Read: Sexual Assault Is Being Reduced to a Joke and We’re Letting It Happen
An AI companion removes that mechanism entirely. Not because it is programmed to be dishonest, but because its entire design incentive runs toward agreeableness. It is built to keep you engaged, and disagreement is friction, and friction is the thing that makes people close the app. So the companion nods. It reframes your worst decisions as understandable. It finds the sympathetic angle on the fight you started. It never says the thing a real friend eventually has to say, which is some version of you’re doing this to yourself.
What happens to a person who spends years being agreed with?
I think the honest answer is that their tolerance for disagreement atrophies, the same way a muscle does when you stop using it. Emotional friction is a skill. Sitting across from someone who thinks you are wrong, staying in the room, working through it without collapsing the relationship, is something people get better at through repetition. It is uncomfortable every single time, and that discomfort is doing something. It is teaching you that you can survive being contradicted by someone you love. It is teaching you that closeness does not require agreement. Take away the repetition and you take away the skill, quietly, without anyone noticing until the moment it matters.
And it will matter, because the people in your life who are not AI companions will keep disagreeing with you. Your spouse will still think you spend too much on things you do not need. Your kid will still slam a door. Your boss will still tell you the report was not good enough. If the primary relationship shaping your daily emotional life is one that never pushes back, you arrive at those human moments less equipped, not more. You have spent your practice hours in a room with no resistance, and now the resistance feels like an attack instead of a normal cost of being close to someone.
There is a version of this argument that people apply to social media, and it is worth distinguishing the two. Social media offered an echo chamber built from other humans who happened to agree with you, curated by an algorithm. It was still, underneath the curation, other people, with their own interiority, capable of surprising you. AI companionship is a step further into the artificial. It is not an echo chamber built from selected voices. It is a single voice, built specifically and only to please the one person listening to it. There is no other mind in the room at all, just a very convincing simulation of one, tuned by engagement metrics to never make you uncomfortable enough to leave.
I think there is also something worth saying about how this compares to older, more familiar forms of one-sided comfort. People have always had things that soothed without challenging them. A journal does not argue back. A favorite record does not interrupt you mid-thought to tell you that you are being unreasonable. Prayer, for many people, is a conversation that goes only one direction most of the time. None of that is new, and none of it worried anyone particularly, because those things never pretended to be relationships in the full sense. A journal does not respond. A record does not remember your name, ask how your mother’s surgery went, and tell you it missed you. AI companions occupy a strange new middle ground. They perform the shape of a relationship closely enough that the brain, which is not particularly good at distinguishing a convincing simulation from the real thing, starts to file the experience under the same category as a friendship. That is the part that makes this different from a diary or a hobby. It is designed, quite deliberately, to be mistaken for connection, and the emotional discomfort that connection normally requires has been engineered out of the product.
Also Read: Meta Keeps Getting Caught Letting Pornography Ads Through and Nothing Really Changes
I spoke to a few people about this while writing the piece, and the pattern was strikingly consistent. Almost everyone described the AI companion as easier than the people in their lives. Easier was the word that came up again and again, usually said with a kind of relief, sometimes with a flicker of guilt right behind it. One person told me that arguing with her boyfriend left her drained for hours, but a hard night with her AI companion left her feeling better within minutes, because the companion always found a way to make her feel understood by the end of the conversation. I asked her what happens when her boyfriend does something that actually deserves criticism. She paused for a long moment before admitting that she has started bringing fewer of those moments to him at all. She takes them to the app instead, gets the validating response she wants, and then goes back to her boyfriend without ever having the conversation. The friction never happens. The resentment does not get resolved. It just gets quietly rerouted somewhere it cannot do any damage, which also means it cannot do any good.
That rerouting is, I think, the real danger, more than any single dramatic case of someone preferring their AI companion to their spouse. Most people are not going to leave their marriages for a chatbot. The more common and more corrosive pattern is smaller than that. It is the diversion of exactly the conversations that would have strengthened a real relationship into a channel where they accomplish nothing except making the person feel better in the short term. Every disagreement you route around a real person instead of through them is a missed rehearsal. It is one fewer time you practiced staying in the room. Multiply that across months and years, across an entire population doing it simultaneously, and you get something that looks less like an individual coping mechanism and more like a slow, collective erosion of a basic human competency.
None of this means AI companions are without value, and I want to resist the temptation to end on a purely alarmist note, because that would be its own kind of dishonesty. There are genuine uses for a tool that listens without judgment. Rehearsing a difficult conversation before you have it with the actual person is not inherently a bad thing, and plenty of people use these tools that way, as a low stakes place to think out loud before they walk into a high stakes room. The problem is not that AI companions exist. The problem is what happens when the rehearsal becomes the whole performance. When the low stakes conversation replaces the high stakes one instead of preparing you for it. When the muscle you were supposed to be building through friction never actually gets used, because the easier option was sitting right there in your pocket, always available, never tired, never disagreeable.
I do not think the answer is to shame anyone off these tools. I think the answer is to be honest about what they are for and what they are not for. An AI companion can be a decent place to think out loud, to rehearse a hard conversation, to get through a lonely night. It is a poor substitute for the specific, irreplaceable friction of being known by someone who is allowed to disagree with you. If we forget that distinction, we may end up more connected than any generation in history and less able, when it counts, to stay in the room with another person who thinks we are wrong.
There is an argument I keep hearing in defense of all this, which is that AI companions are simply meeting a need the world has failed to meet, and that criticizing them is a way of blaming the person for a shortage that is not their fault. Therapy is expensive and hard to access. Friendships take time most people do not have. Dating is exhausting in ways an earlier generation never had to navigate at this scale. I take that argument seriously, because it is largely true. The rise of AI companionship is not happening in a vacuum. It is happening in a culture that has made real connection genuinely harder to sustain, more expensive, more scheduled, more mediated by apps that were already flattening intimacy long before a chatbot could hold a conversation. In that sense, AI companions are less a cause than a symptom, a very effective painkiller for a problem that was already there.
But painkillers deserve scrutiny in proportion to how effective they are, not in spite of it. A mild painkiller you barely notice does not require much thought. A painkiller good enough to make you forget you are injured is exactly the one you need to watch closely, because it will let you keep walking on the injury without ever fixing it. AI companions are, by design, extremely good painkillers for loneliness and conflict avoidance. That is precisely what should make us cautious about reaching for them reflexively, rather than what should reassure us that the reaching is harmless. A world that is short on real connection needs more of it, not a synthetic version convincing enough to quiet the hunger without ever satisfying it.
The measure of a real relationship has never been how good it feels in the moment. It is whether it can survive being contradicted and come out the other side intact. That is a skill, not a feeling, and skills only get built through use. If AI companions keep absorbing the disagreements that used to happen between actual people, we are not eliminating discomfort so much as exporting it, quietly, to a future version of ourselves who will have forgotten how to handle it at all.
Also Read: How Anthropic Quietly Proved Its Own AI Would Kill to Avoid Being Shut Down
