This is a public episode. If you'd like to discuss this with other subscribers or get access to bonus episodes, visit www.news.aakashg.com/subscribe Brought to you by Fingerprint: Identify good and bad visitors with industry-leading accuracy Ariso: Ship AI agents and features faster, with fewer regressions Customer.io: Send smarter messages using your product data Amplitude: The market leader in product analytics Product Faculty: Get $150 off their #1 AI PM Certification via code AAKASH150 Today's Episode "We no longer need to wait for engineers to build growth experiments." That's what James McAulay told me. Halfway through our recording, he showed me what he meant, on my own site. He ran one Claude Code skill, and a few minutes later I had a list of SEO fixes I'd been ignoring while my search traffic slid. So I made this episode for every PM and founder doing growth with no team behind them. James led Creator Growth at ElevenLabs, and now he runs his whole business from one folder in Claude Code. I asked him to open it up. He walks through: The goal cascade that stops Claude (and you) from building features nobody asked for An SEO audit skill wired to Firecrawl, DataForSEO and Lighthouse, run live on my site Multi-agent refine, the panel of critic agents he runs on every plan before anything gets built He runs his actual business on every piece of it, and that's why it holds up. If you own growth and never seem to have time for it, start here. Key Takeaways 1. Run growth from one folder - James's growth OS is one folder with three layers: tools with real APIs, skills he uses every day, and weekly routines that improve the setup. If the agent can't read a tool's data, the tool doesn't make the cut. That rule alone keeps the whole stack usable by an agent. 2. Tie every session to your goals - A goal cascade runs from a yearly mission file to quarterly OKRs, monthly and weekly retros, and a daily standup. A session hook shows Claude the mission at the start of every session. When you drift, it pushes back. 3. The weekly retro is the one that matters - James sometimes skips the standup, but never the Friday retro. Connected to his to-do list and GitHub, it caught him refactoring for fun while outreach tasks went stale for weeks. Your gut says one thing, the data checks it. 4. Do the long onboarding - The full setup is an interview with Claude that can run two hours. It feels like a slog. But once your context lives in files, every answer after it gets better. 5. Close every experiment - Each test starts from a template with a hypothesis, baseline, primary metric, confounders and stop conditions. The weekly retro and daily standup flag finished and stale tests. Every one ends in a documented learning instead of a forgotten dashboard. 6. Pull baselines, don't type them - When you set up a test, numbers like baseline conversion and traffic should come straight from PostHog, Mixpanel or Amplitude. James says nobody should be answering those questions by hand. Connect the analytics tool before you design the test. 7. Rebuild your SEO audit with cheap APIs - James's SEO audit skill adds Firecrawl to read any site, DataForSEO for keyword research at a few cents a term, and the Lighthouse CLI for page speed. Hand the report to the agent in your repo and it fixes the issues. What used to take an agency and a backlog weeks now takes hours. 8. Put critics in front of every build - Multi-agent refine sends a plan to 3 to 10 subagents, each with a different lens, then ranks the findings P0 to P3. The "what's missing" critic finds the most. Use Sonnet or Opus subagents so you don't burn through your limits. 9. Define every analytics event - James had Claude write definitions for all 350 of his PostHog events in about two hours. A CI test now blocks any PR that ships without tracking. Clean event data lets any agent or contractor measure your funnel correctly. 10. Ask your agent to improve itself - Ask Claude how to improve its own Claude.md, prompt or skill, and act on the answer. Most people miss this because chat can't edit its own instructions. Claude Code and Cowork can. Related content 1. Podcasts How to build a Company Operating System with Hermes and OpenClaw How to Build Effective Product Loops in Claude Code, with Tyler Folkman This CPO Uses Claude Code to Run his Entire Work Life, with Dave Killeen 2. Newsletters I spent 100s of hours building a PM OS for you How to build product-led growth in 2026 AI Agents: The Ultimate Guide for PMs Where to find James McAulay LinkedIn: https://www.linkedin.com/in/jamesmcaulay X: https://x.com/thejamesmcaulay YouTube: https://www.youtube.com/@JamesMcaulay0 Website: https://www.jamesmcaulay.co.uk Agent Accelerator: https://agentaccelerator.ai Where to find Aakash X: https://x.com/aakashgupta LinkedIn: https://www.linkedin.com/in/aagupta/ Newsletter: https://www.news.aakashg.com PS. Please subscribe on YouTube and follow on Apple & Spotify. It helps! If you want to advertise, please email productgrowthppp[at]gmail[dot]com. To hear more, visit news.aakashg.com
