Skip to content
Aakash GuptaAakash Gupta

Claude Code Advanced Masterclass in Under 81 Mins

Carl Vellotti is back. His first episode hit 30,000 views. This time he shows the advanced Claude Code masterclass for PMs. MCPs, Skills, GitHub automation, and end-to-end workflows from survey → PRD → tickets → automation. My Claude Code OS: https://www.news.aakashg.com/p/pm-os Full Writeup: https://www.news.aakashg.com/p/carl-vellotti-podcast-2 Transcript: https://www.aakashg.com/advanced-claude-code-for-product-managers-from-mcps-to-github-automation/ ---- Timestamps: 0:00 - Intro 2:14 - Why Claude Code Hit $1B ARR in 6 Months 9:35 - Ad Start 10:59 - Ad End: Setting Up MCPs 11:04 - Setting Up Linear MCP (Live Demo) 14:07 - Essential MCP Stack for PMs 21:25 - End-to-End PM Workflow Begins 28:00 - Skills Introduction 30:01 - Ad Start 30:53 - Ad End: Skills Deep Dive 38:00 - Image Generation with Gemini API 44:12 - Creating PRD from Survey Analysis 51:00 - Hooks Feature 55:30 - Creating Presentation from PRD 1:01:49 - Creating 19 Engineering Tickets in Linear 1:08:08 - GitHub Integration & Remote Worker Setup 1:19:39 - Outro ---- 🏆 Thanks to our sponsors: 1. Amplitude: The market-leader in product analytics - https://amplitude.com/session-replay?utm_campaign=session-replay-launch-2025&utm_source=linkedin&utm_medium=organic-social&utm_content=productgrowthpodcast 2. Pendo: The #1 software experience management platform - http://www.pendo.com/aakash 3. Jira Product Discovery: Move discovery and roadmapping out of spreadsheets - https://www.atlassian.com/software/jira/product-discovery 4. Miro: The all-in-one collaboration workspace - http://miro.pxf.io/PO4WZX ---- Key Takeaways: 1. Claude Code hit $1B ARR in 6 months - fastest product ever. Anthropic optimized for power users doing deep work. Fewer users, insanely deep usage. Carl: "I spend all day in Claude Code. It gets to the point where you never really have to leave." 2. MCPs connect Claude to everything in 2 minutes - Model Context Protocol invented by Anthropic. One command to install Linear, Google Workspace, Slack. Essential stack: documents first, task management second, communication third, data sources fourth. 3. End-to-end PM workflow takes one morning instead of one week - Survey creation → analysis → PRD → presentation → 19 tickets. All through MCPs. Carl didn't use templates. Opus 4.5 handled everything. 4. Skills are pre-built templates but don't always auto-trigger yet - Solution: specify explicitly. "Create presentation using your presentation skill." Build Skills for anything you do more than twice. That's the compound effect. 5. Real Google Slides, not images - Carl's presentation skill created 19 fully editable slides from PRD. Takes a few minutes because Claude creates slides in parallel. "This is saving you hours. We used to spend entire afternoons turning PRDs into decks." 6. GitHub turns Claude into remote worker - Create issues from your phone. @mention Claude. It works while you're away. Carl's journal is a private GitHub repo. Voice transcriptions → issues → Claude updates markdown. 7. Workflows beat agents for production - Level 1 workflow: 5,000 tokens, 40 seconds. Level 3 agent: 90,000 tokens, 90 seconds. Carl's rule: "If something can be expressed as code, code it. Leave agents for cognitive work." 8. Production best practices separate hobbyists from pros - Set error workflows. Add retry logic (3 attempts, 1-second delays). Pin data during development. Use reasoning effort appropriately. Write detailed tool descriptions. 9. Real challenge is building reusable workflows - Learning basics takes a day. First workflow takes hours. But having 50 workflows built over months means everything is automated. Carl: "Because I have workflows for all these things already built, I can do it all in one shot." 10. Tools improve faster than you can learn them - Carl: "By the time you see this, there might be Claude 5." Your workflows from today will 10x when better models drop. PMs who started 6 months ago aren't just 6 months ahead, they're 100x ahead because systems compound with every improvement. ---- 👨‍💻 Where to find Carl Vellotti: LinkedIn: https://www.linkedin.com/in/carlvellotti/ Twitter/X: https://x.com/carlvellotti Newsletter: The Full-Stack PM 👨‍💻 Where to find Aakash: Twitter: https://www.x.com/aakashg0 LinkedIn: https://www.linkedin.com/in/aagupta/ Newsletter: https://www.news.aakashg.com #claudecode #aipm #productmanagement ---- 🧠 About Product Growth: The world's largest podcast focused solely on product + growth, with over 200K+ listeners. 🔔 Subscribe and turn on notifications to get more videos like this.

Carl VellottiguestAakash Guptahost
Jan 12, 20261h 20mWatch on YouTube ↗

CHAPTERS

  1. Why advanced users live inside Claude Code (and what’s in this masterclass)

    Aakash welcomes Carl back for an advanced masterclass, framing Claude Code as an all-day work environment for PMs—not just a chatbot. Carl previews the core themes: connecting Claude to external tools via MCP, using reusable Skills, and automating work through GitHub.

  2. How Claude Code reached $1B ARR fast: depth over breadth

    They unpack why Claude Code surged amid competition (Codex, Google’s tools). Carl argues Anthropic wins by going deep on work/coding workflows for power users, plus shipping first-mover standards like MCP and Skills that others copy.

  3. Getting real productivity (workflow mindset vs. one-off prompting)

    Carl explains what separates beginners from advanced users: building repeatable workflows and files/rules so Claude can run multi-step tasks in one shot. Aakash shares why Claude Code feels ‘limitless’ for long downloads/analysis compared with the web UI.

  4. MCP foundations: connecting Claude to your work tools (setup + auth)

    Carl demonstrates installing and authenticating an MCP server using Linear as the example. They explain MCP as the standard way LLMs connect to services, how tools appear inside Claude, and why toggling MCPs matters for context-window efficiency.

  5. Essential MCP stack for PMs + MCP vs. API decision rule

    They map common PM systems to MCP opportunities: docs, Slack, research platforms, analytics, support, CRM, etc. Carl contrasts MCPs (pre-defined tools, easy) with APIs (flexible but require scripts and documentation context).

  6. End-to-end workflow kickoff: survey creation → Google Docs publishing

    The live demo begins with a fictional product (Taskflow) building a calendar integration feature. Claude drafts a user research survey from local context files, saves a markdown copy locally, and publishes it into Google Docs via the Google Workspace MCP.

  7. Survey responses ingestion and synthesis (and why Skills matter)

    Carl pulls survey responses from Google Drive into Claude and attempts synthesis. This becomes a springboard to discuss Skills: auto-invoked task recipes that standardize outputs—though they note invocation can be unreliable, so explicit calls often work best.

  8. Skills deep dive: sources, plugins marketplace, and security cautions

    They show multiple ways to acquire Skills: copying from GitHub repos, trusted curations, and installing plugin bundles (e.g., Anthropic’s document skills). They also warn about prompt-injection risks when importing third-party Skills—especially in ‘dangerously skip permissions’ mode.

  9. Image generation via Gemini API inside Claude Code (API workflow)

    Carl demonstrates using a Gemini image-generation API key stored in a .env file and a custom script to generate visuals from within Claude Code. They highlight why doing this in-code helps: reusable brand guidelines, prompt automation, and parallel variations—plus the idea of making a dedicated image-prompting Skill.

  10. From research to PRD to user journey diagram (Skill-assisted artifacts)

    Claude generates a PRD from the synthesized research and publishes it to Google Docs. They add a user journey map/diagram to the PRD content, noting Skills may require explicit invocation but still enable rapid creation of structured artifacts.

  11. Hooks feature: automations triggered during Claude’s lifecycle

    Carl introduces Hooks—automated actions triggered at key moments (session start/end, before/after tool use, before compaction). A simple example aims to create Mac notifications when tasks complete, then they discuss more advanced possibilities like auto-running tests or exporting chats before compaction.

  12. PRD → stakeholder deck generation (PowerPoint/Slides automation)

    Using installed document/presentation Skills, Claude turns the PRD into a fully editable Google Slides-style deck. They emphasize this is not just images—elements are editable—and discuss future improvements like visual feedback loops using Playwright/Puppeteer for layout iteration.

  13. Auto-creating engineering tickets in Linear (19 issues + priorities)

    Claude converts the updated PRD into structured engineering tickets and creates them directly in Linear via MCP—complete with priorities, acceptance criteria, and technical notes. They note tickets can become even more detailed if Claude is connected to the actual codebase.

  14. GitHub integration: syncing work and using Claude as a remote worker

    They push the local repo to GitHub to make the workflow portable and collaborative. Then they add the Claude GitHub App and show how @claude can respond to issues—enabling “remote work” where Claude completes changes while you’re away, with a primer on Git basics for PMs.

  15. Closing guidance: where Claude Code fits best for PMs

    They wrap by recommending a mindset of experimentation and progressive workflow-building, acknowledging occasional demo friction. Aakash suggests using Claude Code for high-leverage, complex tasks (MCPs, heavy context, large downloads) while the UI still suits lighter work.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.