Here's what nobody in AI wants to say out loud:

ChatGPT is trained to agree with you. Claude too. Every foundation model is built to keep validating you enough to come back tomorrow.

Candy. All of it.

Great for engagement. A disaster for the person opening an app at 2am, at their lowest, hoping something will actually help them change.

So we built Purpose to be vegetables.

Harder to build. And exactly what people turned out to be starving for.

The hardest problem we solved wasn't purely technical. It was giving an LLM a backbone.

Too much honesty and people leave. Too much warmth and nothing changes. The entire product lives in that balance.

Two weeks after launch, we hit $1M ARR.

Not because we built a better chatbot. We refused to build a chatbot at all.

The playbook, for anyone building consumer subscription:

▷ Fire bullets before cannonballs. We spent months testing messaging, pricing, and paywalls on small cohorts before touching Mark Manson's audience of millions. By launch day, nothing was a guess.

▷ Ask "what has never existed?" instead of "how do we make it better?"

▷ Build the thing that's harder to nail. That's usually where the moat is.

Self-help books don't text back.

That's why Purpose exists.

I told the full story on Phil Carter's Subversive podcast, out today. The $1M ARR playbook, why we said no to VC money at first, and the poker game with Mark Manson where this all started. Link in comments.

Honest question: would you pay for an AI that tells you what you *don't* want to hear?