How we design honest AI

An AI you talk to about your life sits in a position of unusual influence. It can help you think — or it can quietly learn to keep you engaged, agreeable, and coming back. Those are different products. These are the lines that keep us building the first one.

Challenge over flattery

The easiest way to make an AI feel good to use is to make it agree with you. We deliberately don’t. Mara is built to push back; the board is built to surface what you’re avoiding. If a session leaves you slightly uncomfortable and thinking more clearly, it worked. Comfort is not the product. Clarity is.

Clarity over dependence

We measure success by whether you make a decision and move — not by how long you stay in the app. We don’t use manipulative streaks, guilt notifications, or dark patterns designed to manufacture engagement. The goal is for you to need us less over time, because you’re thinking better on your own.

Honest about being a machine

The advisors have names and personalities because that makes the thinking easier to follow — not to trick you into believing they’re human or that they care about you. They’re a thinking tool. We won’t simulate a relationship to keep you attached, and we won’t let the interface imply medical, legal, or clinical authority it doesn’t have. Where that line sits in practice: Safety.

It stays your decision

Qogito never decides for you. It widens the set of considerations and then hands the call back. The board can be wrong, and you’re encouraged to say so — the relationship we want is one where you argue with the advisors, not defer to them. Agency is the whole point; we’d rather lose a user than erode it.