Weekly AI Founder Mastermind recap, October 29: 18 founders on the call

The fourth Weekly AI Founder Mastermind, and the first at 60 minutes, brought together 18 founders from New York, Lisbon, London, Amsterdam, India, South Africa, Austin and more (join next week's call with Lounge Pro). Three topics: rebuilding a 20-year-old SaaS platform to be AI-native, letting AI answer questions from your own data, and code review once AI writes hundreds of pull requests a day. The thread through all three: the AI tools work, and the results come from the system you build around them.

• Growth: Lounge passed 2,500 members, about 230 of them founders at a verified $1M+ ARR.
• City ambassadors: Berlin, Lisbon (Oliver Schmalholz hosts), Ankara, Mogadishu, Tokyo, London, Reno and Austin, with Montreal and Cleveland applying. One event a quarter, with about $1,000 per city for food and drinks once sponsors are on board. Want to host or sponsor? Message Ryan.

Keith Pelatowski (Chief Executive Officer, Points North) runs a profitable, 70-person company whose 20-year-old platform now keeps AI out of its users' workflows. About 15 engineers have Claude Code; 2 or 3 use it daily. The goal: an AI-native platform by April or May.

• Ron Laughton (Founder & CEO, ReviewInc): three developers on $100-a-month Claude Code plans rebuilt a 15-year-old portal (Bootstrap 5, Section 508 accessibility, site-wide search, an AI help assistant) in about three weeks instead of six months. Secrets went into Azure Key Vault first, and Claude got a schema dump, never the database.
• Antonio Figueroa (Founder and CEO, Raken AI Labs) modernizes version by version: write down the business rules nobody documented, prototype with the customer in short cycles, and build the back end in parallel with Claude skills. Raken's gross margin went from 40% to 67% in six months.
• Besnik Vrellaku (CEO & Founder, Salesflow) rebuilt on a new agentic stack over 8 months and kept the infrastructure that worked. If a full rebuild is too much, start with one module.
• Rick Cancelliere (Founder and CEO, PracticeBeat) put a number on the problem (60 to 90 day implementations, about $25,000 a month in delayed revenue each), set a one-day go-live goal and tied bonuses to it.
• Chason Hecht (Retensa) would build in-house if you have runway, for the team's pride of ownership. Brook Schaaf (Co-Founder and CEO, FMTC) said guardrails, not models, are where things break now.

Lessons learned

• Tie the rebuild to a business number and to the team's bonuses.
• Document the unwritten rules first, then rebuild in short cycles, one module at a time if needed.
• Secure it before you start: secrets out of the code, a schema instead of database access.

Ron wants ReviewInc customers to ask their 100+ reports questions in plain English. MCP with a temporary token, or server-to-server?

• Antonio built this for a telecom with a petabyte data lake. What fixed the hallucinations was a metadata layer: what every KPI means, where it lives and what people will ask. Starburst and Qlik Answers now build much of it for you.
• Ricardo Ghekiere (Founder, Patricia.app) maps where the data lives instead of importing it, reviews every failed prompt each Monday, and plans for up to 50 rounds per use case.
• Ron's plan: an MCP layer filtered to each user's access, with PHI and PII removed before the model sees anything.

Lessons learned

• Build the metadata layer first. MCP is the connector, not the solution.
• Filter by permission and sanitize sensitive data before the model touches it.

• Ricardo outgrew CodeRabbit, built his own review tool and batches PRs into "trains" so the merge queue doesn't jam. Costs fell from about $4,000 to $1,000-$1,500 a month.
• Cristian Frunze (Founder, Ken), #4 on Lounge's GitHub leaderboard, moved CI/CD to bare-metal servers. A classifier routes each PR by size: an open-source model or Grok for small changes, Opus for medium, the top OpenAI models plus Opus for the largest, with 3 to 5 review rounds. It took two days to build.
• Ken's numbers: in 30 days, a team of two merged 4,666 PRs on 371 billion Claude Code tokens, about $172,600 at API list prices, paid with accelerator credits. That is about $15 per commit, the benchmark Chason suggested everyone track.
• Ultracode mode: Cristian plans a module as Linear tickets, then lets Claude Code run 10 to 50 sub-agents on it overnight.

Lessons learned

• Past about 50 PRs a day, build your own review pipeline and route reviews by size.
• Know your cost per commit.

Andrei Stefan (CEO, Vitals), Richard Turner (President & Founder, Convergence Data), Joey Trampush (Founder & Chief Neuropsychologist, brainworkup), Kalp Jain (Developer, Independent), and from the SaasRise team David Trachsel and Lee Radloff.

Thursday, November 5 at 11 AM CT, 60 minutes on Zoom. Add a topic on the mastermind page and vote on what we cover. Ideas from today: monetizing AI inside your app, and AI pricing as teams move to open-weight models.