Weekly AI Founder Mastermind recap, October 22: 17 founders on the call

Two topics, 30 minutes, and a room of founders running AI and SaaS companies, most at $1M+ ARR. The full recording is here, so you can watch the whole thing. The headline: AI go-to-market is moving out of SaaS tools and into agents and code, and the strongest R&D teams on the call now ship most of their code through AI, with the hard problems moving to review, CI/CD and judgment.

• Rick Cancelliere (Founder and CEO, PracticeBeat) runs separate agents and co-pilots for inbound, outbound and partner channels, built on tools like Replit and Cowork. Agentic outreach is replacing a stack of expensive SaaS tools, and partners get daily, weekly and monthly agent-written briefs on which prospects are worth an assessment.
• Tom Regan (Founder & CEO, ArtemisGTM.ai) builds his GTM in code: his own agents in Claude Code plus a set of skills he runs every day. His three picks: DeepLine, a headless GTM layer for enrichment, routing and CRM hygiene that runs from the terminal or Claude Code; Amplemarket and its AI assistant Duo for SDR and BDR teams working calls, LinkedIn and email together; and Clay, which he rates but no longer needs because his Claude Code workflows do the job.
• Steve Behrisch (President and CEO, OnRes) is re-engineering lead generation around agents, with Grok built into the process.
• Louis Hendriks (Founder & CEO, Global Value Web) is testing two AI SEO firms, because his buyers now search through AI assistants and the company has to show up there.
• Ryan Allis (CEO, Lounge.ai & SaasRise) shared how Lounge grew: Claude researches founders and developers from GitHub, Product Hunt, Hacker News, Y Combinator and accelerator portfolios through the Moltsets and Apollo APIs, builds each person a profile, and invites them to claim it, with one follow-up. About 40,000 invites in two weeks converted at 3.4%, about 1,300 of the first 1,700 members. Built in-house, not on Instantly.

Lessons learned on AI GTM

• Build the workflow before you buy the tool. Several of the best setups on the call are agents and Claude Code skills, not seats in another SaaS product.
• Use AI to do the research, then make the first touch personal. A pre-built profile someone can claim beats a generic sequence.
• Give partners and channels their own agents. Regular, specific briefs on why a prospect is ready move more deals than a shared spreadsheet.
• Plan for AI search. If buyers ask ChatGPT or Claude for a vendor, being cited there is the new page one.
• The research-heavy invite playbook gets more valuable as ACV goes up: mid-market and enterprise sellers can afford deeper research per account.

About a year ago, roughly 10% of a similar room raised a hand for using AI in development. Today it was about 80%.

• Ricardo Ghekiere (Founder, Patricia.app) runs about 90% of the code with no human in the loop. Customers report bugs to a Slack bot, most bugs fix themselves within four hours, and his CTO's own bot reads the company Slack overnight and ships fixes for anything people complain about. Feature requests still need his approval. The bottleneck moved to CI/CD: the merge queue took 20 to 30 minutes per PR until they rebuilt it to about three, and they replaced Code Rabbit with their own review system.
• Andrei Stefan (CEO, Vitals) is shipping a self-improving dashboard. An AI reads Amplitude usage every day, removes widgets nobody uses, and writes the code for new ones to try. Next is fake-door testing: put an idea in the product, and if people click it, the product builds it. His view on autonomy is a confidence threshold for going live without a human, lowered as the system earns trust.
• Chason Hecht (Retensa) built a strict pipeline: Claude Code writes, the team reviews, automated AI testers run in a test environment, then staging, then a mandatory human approval before production. With sensitive customer data, AI does not make the final call. He expects five to eight times the output from a smaller team.
• Steve Behrisch runs his team out of Cursor projects connected to their Git repos: ChatGPT acts as project manager, Claude does the work, and they pick the model per task, Grok for simple jobs and Claude Opus for the heavy lifting.
• Ron Laughton (Founder & CEO, ReviewInc) reviews Claude's output by hand for hours a day and gave the warning of the call: Claude Code is "book smart but not street smart." It writes verbose code, rebuilds routines that already exist, and adds packages and cookies that aren't needed.

Lessons learned on AI R&D

• When AI writes most of the code, the constraint becomes CI/CD. Fix your merge queue and review step before you add more agents.
• Close the loop from customers to code. Bug reports, Slack threads and product analytics can all feed the agents directly.
• Decide where humans stay in the loop on purpose: production deploys, sensitive data, and new features, at least until the system proves itself.
• Review the code, especially on a legacy codebase. Without a human pointing it at existing routines, AI adds duplication that compounds over ten years of code.
• Match the model to the task. Cheaper models for simple work, the strongest model for architecture and hard bugs.

The group voted to make the call 60 minutes, starting Thursday, October 29 at 11 AM CT. Bring a topic and vote on what we cover.

The Weekly AI Founder Mastermind is for Lounge Pro members and for founders with $1M+ ARR that we can verify through Apollo or Moltsets. If you got value from this recording, the live call is where you can bring your own problem to the room.