Ghost

The Thesis

Ghost came out of the same period as Myeverly, during the broader surge of interest in AI companions and always-available AI chat. The angle was privacy: an AI chat experience built around zero-footprint conversations — no account, no training on your data, messages that could disappear on a timer. It’s a real, well-understood need; people want to think out loud to something without a permanent record of it.

What I Built

I got far enough to validate the core mechanics of a privacy-first chat experience worth building — before I’d properly checked two things that should have come first.

What Happened

First, the regulatory and mental-health risk around AI-therapy and AI-companion-adjacent products is real, and it’s the kind of category where scrutiny only tends to increase over time, not decrease. That’s a hard thing to build a durable product on top of.

Second: there wasn’t a real product to sell. ChatGPT and Claude already ship private and ephemeral chat modes natively, so a standalone app pitching “private AI chat” as its whole differentiator doesn’t hold up against a free feature already built into the incumbents. I’d validated the mechanics before I’d done that basic competitive check — the wrong order.

What I Learned

Check regulatory exposure and competitive differentiation before you build the mechanics, not after — it’s a cheaper mistake to catch early.

• A sensitive category deserves the assumption that scrutiny is only going to increase, not decrease.
• When a big incumbent ships your core differentiator for free, that’s a market signal, not an obstacle to design around.
• The technical opportunity being real was never the right bar — the right bar is whether the category and the competitive position both hold up.