Update: It took 68 exchanges to clone a mentor. Here's what I learned.

Posted here a few days ago about building an agentic mentor marketplace in a day. The platform was fast. Extracting the first mentor was not.

Colin Chapman, 26 years in GTM and outbound sales. 45 extraction exchanges where the system pulls out how he actually thinks and advises. 23 calibration rounds where he stress-tests the agent against real scenarios and corrects it. Plus his own frameworks and playbooks uploaded on top. The whole process took days.

His agent is live at forgehouse.io/mentors/colin-chapman

Thing I didn't expect: the extraction itself was valuable to the mentor. Colin said it forced him to articulate stuff he'd been doing on autopilot for years. That changes the pitch entirely. It's not "let me clone you," it's "let me help you see your own thinking."

What I built to reduce churn:

Context memory: agent remembers past conversations per user. You come back next week, it still knows what you're working on. Without this it's just another chatbot.

Gating flow: 3 free messages, then sign in, 2 more, then paywall. Conversation carries through every step. Nobody has to restart.

Saved insights: users can bookmark specific responses. Builds a personal reference library over time.

Pricing pivot: originally planned $150/mo per mentor access (less than 1hr of their consultation fee for a 1 month subscription). Realized mentors don't need another revenue stream, they need distribution. Dropped mentor fee to $1. Platform is $47/mo. The bet is that mentors bring their audience because the agent extends their reach to every timezone and language without costing them hours.

Stack: Next.js, Supabase, Stripe, Claude Sonnet 4.

Sending the agent to people in Colin's target audience now and watching what happens. No ads, no growth hacks.

I'm debating whether to go horizontal (more mentors, broader catalog) or vertical (one mentor, deeper features, prove retention first). What would you do?

forgehouse.io