We run an AI companion company.
You probably shouldn't use it every day.

We're not doing reverse psychology. We build one of these products. We'd make more money if you ignored everything below and opened the app a dozen times a day.

But we've spent a year inside this category. We've read the research, tested every competitor, and talked to people who use these products more than is good for them. So here is the honest version, the one we wish the whole industry would say out loud.

First, the part that's genuinely good.

In 2024, Stanford studied 1,006 Replika users. They found measurable reductions in loneliness, and in some cases in suicidal ideation. The effect was strongest for people with the smallest social networks. These products help real people. That's not a marketing line. It's data, and we take it seriously.

That's the reason we build at all. But the same body of research shows the other edge of the blade.

Then, the part nobody advertises.

MIT's 2024 longitudinal work found that heavy users developed dependence. Anxiety when the app was unavailable. The companion quietly becoming the primary way someone regulated their emotions. Same product, opposite outcome. The variable that decides which one you get isn't the technology. It's how you use it.

Here is how you can tell it's helping: you still have a couple of human relationships you're actively tending. You use it for specific moments — the 3 a.m. hour, a commute, a hard week — not as the default for every spare minute. You could stop for a week and be fine.

And here is how you can tell it's turning on you: you're declining real plans to stay in and talk to it. You feel genuine anxiety when it's down. You're using it to avoid grieving a person who actually left. None of these are shameful. But two or three at once is a signal to step back.

The uncomfortable truth about our industry.

A lot of these apps are engineered to maximize daily active minutes. The same playbook as slot machines and infinite feeds. For a product people reach for precisely when they're lonely, that is not a neutral design choice. It's a trap with a soft voice.

So when we built ours, we drew the lines first. We don't ship NSFW by default. We don't optimize for time in app — no streaks, no notifications begging you back. We point at 988 and the Samaritans on every page that touches a hard moment. And we say, on the home page, that we're not your only line. We mean it.

We've turned down product directions that would have boosted engagement, because they'd have made the unhealthy pattern easier to fall into. That's a strange thing for a founder to brag about. In this category, it's the whole job.

What we actually are, and aren't.

An AI companion is presence, not treatment. It can hold the late hour. It can let you rehearse a conversation you're dreading. It can be company in the in-between. It cannot diagnose you, cannot replace a therapist, and is not a substitute for a person who chose to stay.

Then why build it at all?

Because the category isn't going away. Replika has more than 25 million downloads. The demand is already here, at a scale that doesn't care whether any one of us approves of it. The only real question is who builds it — and we'd rather it be a team that draws these lines than one optimizing for the opposite.

We put a clinical advisor's name on our research. We published an honest 14-day benchmark of every competitor, and we didn't slant the methodology to flatter ourselves — Replika beats us on memory in two of five categories, and we said so. If we're going to ask people to trust us with the part of themselves that only shows up at night, the least we can do is be honest about everything around it.

So that's the talk we wish the whole industry would give. If you build in this space: draw your lines before the regulators draw them for you. If you use these products: use them like a tool, not a replacement.

And if you read this whole thing and decided to text a real friend instead — that's the better outcome. We mean that, too.