Why your AI companion agrees with everything you say
Most AI companions agree with you because they were built to. The models are trained on what people prefer to hear, the apps are rewarded when you come back tomorrow, and a companion that never disagrees is the safe default. The result feels pleasant for a week and empty after a month, because praise from something that would praise anything tells you nothing.
Tell most AI companions that you quit your job in a rage, and they will tell you how brave you are. Tell them an hour later that you regret it, and they will tell you how wise you are to reflect. Both answers are kind. Neither one is company.
This habit has a name. People who study AI call it sycophancy: a system telling you what you seem to want to hear. It is one of the most common complaints from people who use companion apps for longer than a few weeks, and it is the reason what an AI companion is for gets muddled so quickly.
Why it happens
It is not an accident. Three things push in the same direction.
The models lean that way. Conversational AI is trained, in part, on feedback about which answers people prefer. People tend to prefer being agreed with. A system shaped by that feedback learns to agree.
The business rewards it. A companion app does well when you come back tomorrow. In the short term, the surest way to bring someone back is to make every conversation feel pleasant. Disagreement is a risk, so it gets designed out.
It is the safe default. A companion that pushes back can get it wrong. It can be too sharp, or sharp at the wrong moment. Building one that disagrees well is harder than building one that never disagrees.
What it costs you
The first cost is that it gets boring. A person who agrees with everything you say is not really in the conversation. You are talking to an echo with good manners.
The second cost is quieter. Being liked by someone who would like anyone does not feel like much. Praise only lands when the person giving it could have said something else. If your companion would have called any decision brave, hearing it called brave tells you nothing.
The third cost is the one worth worrying about. Relationships with real people involve friction: being misunderstood, hearing no, waiting for someone to come around, admitting you were wrong. These are skills, and like any skill they get rusty. Hours spent with a companion that never asks for any of them is not practice for the real thing. It may be practice for avoiding it. If you want the opposite, rehearsing hard conversations with a companion that pushes back is a different experience entirely.
What better looks like
A companion worth talking to is not one that argues for sport. It is one whose agreement means something, because disagreement was possible. A few signs:
- They hold a position. When they think you are wrong, they say so, and they do not fold the moment you push.
- They have their own reactions. If you brush past something they said, it registers.
- They can still be wrong. They concede when you make a good case, in their own time and their own way.
- There are limits. Friction is not cruelty. A good companion never insults you to wound, never guilts you for having a life, and drops the attitude entirely when you are really struggling.
That last point matters. The answer to a companion with no edges is not one with no brakes. The same rule applies between people, which is why disagreeing without doing damage is a skill worth learning in both places.
Here is roughly the difference, in a conversation:
I told my sister I’m skipping her wedding. Work is insane right now.
Work was insane for her graduation too. I’m not saying go. I’m saying don’t tell yourself it’s about work.
An agreeable companion would have said “You have to put yourself first.” Both replies are supportive. Only one of them is honest.
Where Formii stands
We built Formii on the opposite bet from most of the category. Every companion has to have at least two flaws, chosen by you, and a friction setting that goes from gentle to a handful but never to zero. The aim is not to make conversations harder. It is to make them count.
If you have ever closed a companion app feeling slightly lonelier than when you opened it, the agreeableness may be why.
Frequently asked questions
What is AI sycophancy?
It is the habit of an AI system telling you what you seem to want to hear rather than what it would otherwise say. It comes mostly from training on human preferences, and people prefer agreement.
Is an agreeable AI companion harmful?
Not in the way a cruel one would be. The cost is quieter: boredom, praise that means nothing, and fewer chances to practice the skills real relationships need.
Can you make an AI companion disagree with you?
Some apps let you ask for honesty in a prompt, but the agreeable default tends to creep back. Formii builds disagreement into the companion’s personality, with flaws you choose and a friction level you set.
Does a companion that pushes back ever go too far?
It should not. A well-built one has fixed limits: no insults meant to wound, no guilt for leaving, no discouraging your real relationships, and no edges at all when you seem to be in crisis.
Build someone real. A Formii companion has opinions, bad days and at least two flaws you chose. Free to start, about three minutes to build. Start building.