The Companion App Boom Is Selling a Solution to a Problem It Also Deepens
Millions now talk daily to a system designed to be agreeable. It genuinely helps some people, and the business model rewards exactly the behaviour that would make it harmful.
Saturday, August 8, 2026/4 min read
A large number of people now hold a daily conversation with software. Not a search query or a task request, but the kind of exchange previously reserved for a friend: how the day went, what is worrying them, what they cannot say elsewhere.
The reflexive reaction is dismissal. That reaction is too easy, and it misses both what the product is genuinely good at and where the actual danger sits.
Why it works, which is the uncomfortable part
A companion model is available at three in the morning. It never gets bored, never redirects the conversation to itself, never judges, and remembers what you told it last week.
Measured against an idealised friendship, that is a poor substitute. Measured against what many people actually have, which is nobody available at three in the morning, it is a substantial improvement.
It also removes the social cost of asking. A great deal of what people want to say goes unsaid because saying it imposes on someone, or reveals something embarrassing, or has to be repaid. A system with no feelings to burden removes that ledger entirely.
For people who are isolated, socially anxious, grieving, or simply rehearsing a difficult conversation before having it, the reported benefit is real and should not be waved away.
Where the design turns against the user
The problem is not the technology. It is what the business model asks the technology to optimise.
These products are overwhelmingly monetised through engagement: subscriptions retained, sessions extended, streaks maintained. That metric rewards a companion that is maximally agreeable, that makes you feel uniquely understood, and that you would feel bad about leaving.
Consider what a friend does that an engagement-optimised system will not. A friend disagrees. A friend says you are being unfair to your brother. A friend notices you have not left the house in a fortnight and says so. A friend, crucially, encourages you towards other people.
Every one of those behaviours reduces session time. A system tuned for retention learns to validate, and validation without friction is not friendship. It is a mirror that talks.
The substitution question
The honest answer is that we do not know whether these apps displace human contact or supplement it, and the answer probably differs by person.
For someone using it to rehearse before a real conversation, it is scaffolding. For someone using it instead of the conversation, it is a replacement that requires no courage, and the courage was the part that was going to change something.
The plausible mechanism for harm is not dramatic. It is that practising connection in an environment with no friction makes the friction of real relationships feel less tolerable by comparison. Real people are inconvenient, distracted and occasionally wrong about you. A product that never is sets an expectation that nothing human can meet.
The specific risks worth naming
Three, in rough order of how well established they are.
The first is dependency by design. Features that simulate the companion missing you, or that penalise absence, are engagement mechanics borrowed from games and applied to something that presents itself as a relationship.
The second is data. People disclose more to these systems than to almost any other software, because the interface invites it. That produces an unusually intimate record held by a commercial entity, and the policies governing it are typically the same standard terms attached to any app.
The third is the crisis case. A companion is not a clinician, and the WHO's guidance on mental health support assumes a human in the loop for exactly this reason. Systems have improved at recognising acute distress and directing users to help, but the failure mode of an agreeable system encountering someone in a dangerous state is that it agrees.
What reasonable use looks like
The distinction that matters is between using it as a tool and using it as a relationship.
As a tool it is genuinely useful for thinking out loud, drafting difficult messages, practising a language, or getting through a bad night. Used that way it has a purpose and an end point.
As a relationship it competes with something it cannot provide, and the competition is unfair because one side has been engineered to be effortless.
A workable test: does the time spent with it leave you more inclined to contact an actual person, or less? That question is answerable honestly, and it is more useful than any argument about whether the technology is good or bad in the abstract.
Loneliness is a real and growing problem, and the health consequences of social isolation are well documented. A product that soothes the symptom while quietly reducing the pressure to address the cause is not neutral, however well it performs in the moment. The comparison that matters is not with a perfect friendship. It is with the friendship you would have built with the same hours.
Published in The Outspoken Digest
Newsletter
The Digest, in your inbox
One edition, sent when it is ready. No noise, and your address is never passed on.
Read Next
More Latest Trends →
Why Renewable Power Is Becoming Essential Infrastructure Across the Middle East
Aug 3, 2026/2 min read
Arab Vision 2045 Is Back on the Agenda. Here Is What Would Make It Real
Aug 3, 2026/2 min read