Skip to content
Aakash GuptaAakash Gupta

Complete Course: Claude for PMs (Cowork + Code + Dispatch)

Pawel Huryn tracked 74 Anthropic releases in 52 days and built a PM skills marketplace with 10,000 GitHub stars. In this episode, he walks through his complete AI PM operating system live. Cowork, Claude Code, Dispatch, self-improving knowledge bases, MCP connectors, and remote work across all devices. Full Writeup: https://www.news.aakashg.com/p/ai-pms-guide-to-claude Transcript: https://www.aakashg.com/ai-pms-guide-to-claude/ --- Timestamps: 0:00 - Intro 0:54 - Cowork Power User Demo 3:08 - Anthropic's Shipping Velocity 6:07 - Why Stop Using Chat 7:48 - Ads 09:56 - Cowork vs Code vs Dispatch 18:44 - Skills and MCP Connectors 25:03 - PM Skills Marketplace 29:06 - Strategy Canvas Demo 35:14 - Skill Iteration Cycle 40:46 - Why PMs Need Code 44:43 - Building a Second Brain 56:00 - Self-Improving Knowledge 1:10:00 - Dispatch and Remote Work 1:21:07 - PM Mistakes and Future --- 🏆 Thanks to our sponsors: 1. Arize - AI observability and evals: https://arize.com/?utm_source=aakashgupta&utm_medium=newsletter&utm_campaign=arize_sponsor_ai --- Key Takeaways: 1. Stop using Claude Chat as your default. Cowork accesses real files, connects to Gmail and Slack via MCP, and runs parallel sub-agents. Chat does none of this. 2. Skills are the highest ROI investment. Install marketplace baselines, iterate 5-6 times with specific feedback, and Claude rewrites from first principles until 99% accuracy. 3. Progressive disclosure keeps context clean. Agent reads skill names and descriptions first. Loads full instructions only when the task matches. Hundreds of skills, minimal overhead. 4. Your CLAUDE.md should route, not store. Project structure and pointers only. Domain knowledge lives in separate files the agent loads on demand. 5. Build self-improving knowledge with three types. Rules are confirmed and applied by default. Hypotheses are tracked with evidence. Rejected patterns are kept to avoid retesting. 6. The three-line self-improving prompt works for any domain. Review rules before starting. Apply confirmed rules. Update after feedback. Testing, marketing, strategy, whatever. 7. Claude Code adds explorer view, hooks, subagents, and local MCP scoping. PMs need it once their system grows past 50 files. 8. Every Product Compass infographic was built in Claude Code. HTML generation, component library, iteration through conversation, PNG export. Zero code written by the human. 9. Use Agent Browser from Vercel instead of Chrome MCP. Chrome MCP screenshots every 0.5s and burns $100/hr. Agent Browser uses headless mode and is token-efficient. 10. Dispatch lets you run multiple tasks from your phone. Start an infographic, check emails, analyze competitors. Each runs as a separate thread. Your system works while you live. --- 👨‍💻 Where to find Pawel Huryn: LinkedIn: https://www.linkedin.com/in/pavelhuryn/ X: https://twitter.com/pavolhuryn Product Compass: https://www.productcompass.pm PM Skills: https://github.com/phuryn/pmskills Quadathon (May 9th): 60 spots to build with Claude Code + n8n 👨‍💻 Where to find Aakash: X: https://x.com/aakashgupta LinkedIn: https://www.linkedin.com/in/aakashgupta/ Newsletter: https://www.news.aakashg.com #claudecode #aipm --- 🧠 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.

Aakash GuptahostPawel Hurynguest
May 13, 20261h 32mWatch on YouTube ↗

CHAPTERS

  1. 0:00 – 0:54

    Why Anthropic’s speed signals a “super IC PM” future

    Aakash and Pawel use Anthropic’s rapid release cadence as a lens on where product management is headed. The core idea: companies aren’t just swapping AI into existing steps—they’re redesigning end-to-end workflows around what AI now makes possible, pushing roles closer together.

  2. 0:54 – 3:08

    Why “normal chat” is no longer the default interface for serious work

    Pawel argues standard Claude web chat is fine for quick questions but breaks down for real workflows. The major pain: continuity and portability—when tasks become multi-step, require files, or require switching contexts (coding, exporting, emailing), chat becomes friction.

  3. 3:08 – 6:07

    Tool map for PMs: Chat vs Cowork vs Code vs Dispatch

    They provide an “if this, then that” framework for choosing Claude surfaces. Cowork is for real-file, agentic workflows; Code is for codebases and local execution; Dispatch is for running tasks in the background and from mobile; Chat remains the simplest lightweight option.

  4. 6:07 – 7:48

    Cowork power-user setup: projects, folders, and repeatable workflows

    Pawel shows how he structures Cowork with projects/folders and reusable instructions. The emphasis is on treating Cowork like a workspace for ongoing systems rather than an endless chat thread, enabling repeatable “drop files in → workflow runs” patterns.

  5. 7:48 – 9:56

    Live demo: Cowork organizes invoices, deduplicates, and files by month

    A concrete Cowork example: grant folder access, then ask it to analyze PDFs/images, detect duplicates, create month folders, and move files. The demo illustrates why Cowork beats chat: it operates directly on files and can verify results step-by-step.

  6. 9:56 – 18:44

    Skills + MCP connectors: “USB for agents” and progressive disclosure

    They explain how Cowork/Code become extensible through skills (procedural instructions) and connectors (MCP servers) to external apps. A key concept is progressive disclosure: Claude loads detailed skill instructions only when the skill’s description matches the current task.

  7. 18:44 – 25:03

    PM Skills Marketplace: plugins, workflows, and how to install/use them

    Pawel tours his PM skills repo and how it packages skills into domain plugins (discovery, strategy, GTM, analytics). He demonstrates installing a plugin and using slash commands to force the right skill to run instead of relying on Claude’s default prior knowledge.

  8. 25:03 – 29:06

    How to write great skills (and why iteration is the real ROI)

    They break down what’s inside a skill file: name/description/when-to-use plus detailed step-by-step prompts. The bigger insight is operational: treat skills like products—observe failures, give targeted feedback, and iteratively rewrite the skill to eliminate recurring mistakes.

  9. 29:06 – 35:14

    Strategy Canvas + PPTX demo: McKinsey-level decks in minutes

    Using a product strategy canvas skill plus Anthropic’s PPTX skill, Claude generates a polished multi-slide deck with layouts, icons, metrics, and guardrails. The segment emphasizes that presentation quality is now automatable and directly tied to having strong skills/instructions.

  10. 35:14 – 40:46

    Why PMs still need Claude Code: codebases, file hierarchies, and research automation

    Pawel explains Cowork’s limitations for large repositories and complex multi-file work. Code offers explorer-style navigation, codebase-aware tooling, and enables heavy research + generation workflows (e.g., building viral infographics) where asset management and scripts matter.

  11. 40:46 – 44:43

    Building a “second brain for agents”: self-improving knowledge and content systems

    Pawel describes a system inspired by personal knowledge bases, but optimized for agents: ingest examples, extract patterns, store rules vs hypotheses, and continuously update beliefs. The result is a system that improves outputs (hooks, formats, visuals) based on accumulated evidence.

  12. 44:43 – 56:00

    Minimal viable self-improving setup: keep CLAUDE.md small, route via indexed knowledge

    They warn against bloating CLAUDE.md until it consumes the context window. The preferred architecture: a small CLAUDE.md that explains project purpose and where knowledge lives, plus domain files (indexes/rules/hypotheses) that Claude consults and updates as it works.

  13. 56:00 – 1:10:00

    Browser automation: why not Chrome MCP, and when to use Vercel Agent Browser

    Pawel explains he moved away from Chrome MCP/extension-based browsing due to screenshot-heavy token costs and clunky UX. He recommends Vercel Agent Browser for token-efficient, headless browsing when no API/MCP connector exists, especially for legacy tools or closed platforms.

  14. 1:10:00 – 1:21:07

    Remote work stack: Dispatch vs Web Sessions (and why channels didn’t stick)

    They compare Anthropic’s remote surfaces and how Pawel uses them in real life. Dispatch enables parallel background tasks from mobile but can get messy at scale; Code Web Sessions provide organized, cloud-hosted project access even when the laptop is offline.

  15. 1:21:07 – 1:32:41

    PM pitfalls, what’s underhyped, and when n8n still beats agent-only automation

    They close with lessons learned: the biggest mistake is re-prompting from scratch instead of building learning systems. Pawel argues agents are still underappreciated—but for production-grade automation, deterministic workflows (n8n/code) are safer than “hope the agent follows prompts.”

Get more out of YouTube videos.

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

Add to Chrome