A Mirror, Not a Companion

When we set out to build a dating app that could know you beyond a handful of photos and prompts, our first instinct was an AI that could talk with you, slowly get to know you, and use those conversations to shape your profile and find better matches. On paper, it made perfect sense.

And yet we chose not to build it. This is why.

Three problems stopped us.

The first concern was that, even though we never intended to build an AI companion, a conversation can still turn into one. An AI that's always available, never tired, and never disappointed can pull us into a loop that's hard to step out of. In a four-week trial run by researchers at MIT and OpenAI, the people who relied most heavily on such a chatbot reported the most loneliness and emotional dependence, and the least time spent with other people.¹ We were trying to draw people toward one another. A companion that's always there can become somewhere to retreat instead of a bridge back to the people around us.

The second was what AI does to the truth. It agrees with us. These systems are still evolving, and for now they naturally lean toward whatever we already think. Across eleven leading AI models, researchers found they affirmed people's choices roughly fifty percent more often than another person would — and we trusted them more for it.² They also shape us. When AI helps us put ourselves into words, it subtly influences what we say, and we take on its slant without noticing.³ A system that both agrees with us and shapes us can't show us as we really are.

The third is what we do, all of us, without meaning to. In any conversation, with a person or a machine, we instinctively reach for the self we'd like to be rather than the one we are. We offer our best version, especially in dating, where the whole task is to be chosen. It isn't deception. It's simply something most of us do without realizing it. Put that performance together with a machine inclined to affirm it, and what emerges isn't the truth of a person. It's a version shaped to please. It pleases everyone and reveals no one.

This left us with a realization that the understanding we were after doesn't come from being talked to or talked into. It comes from sitting with ourselves, unhurried and unwatched, and finding the courage to answer questions only we can answer. A machine can't do that work for us, and it shouldn't pretend to by interviewing us into a finished story.

So we built Campfyre the other way around.

Here, AI is a mirror. Not a companion. Not an interviewer.

It doesn't converse you into a profile or supply your words for you. You write freely, honestly, whatever comes to mind — there's no bio to polish, no best self to package. It draws only from what you've already reflected on and chosen to share. It simply helps surface the patterns and values already present in your own words. Nothing nudged. Nothing invented. Nothing that wasn't yours to begin with.

The reflection stays yours. AI simply helps you see yourself more clearly, so one day, someone else can see you too.

Because the point was never to be understood by something that isn't real. It was to understand yourself — and then to be truly known by someone who is.

A mirror, not a companion.

References

1. Fang, C. M., Liu, A. R., Danry, V., Lee, E., Chan, S. W. T., Pataranutaporn, P., Maes, P., Phang, J., Lampe, M., Ahmad, L., & Agarwal, S. (2025). How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use: A Longitudinal Randomized Controlled Study. arXiv:2503.17473. https://arxiv.org/abs/2503.17473

2. Cheng, M., Yu, S., Lee, C., Khadpe, P., Ibrahim, L., & Jurafsky, D. (2026). Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence. Science, 391, eaec8352. https://doi.org/10.1126/science.aec8352

3. Jakesch, M., Bhat, A., Buschek, D., Zalmanson, L., & Naaman, M. (2023). Co-Writing with Opinionated Language Models Affects Users' Views. Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3544548.3581196