Aakash GuptaAakash Gupta

I should be charging $999 for this Claude Code Tutorial

Aakash Gupta and Carl Vellotti on claude Code tutorial for PMs: workflows, agents, PRDs, automation wins.

Aakash GuptahostCarl VellottiguestAakash GuptahostAakash GuptahostAakash GuptahostAakash GuptahostAakash Guptahost
Oct 4, 20251h 37mWatch on YouTube ↗
Terminal-based AI workflow (CLI)Context engineering via folders and filesInstallation and setup (native install)Web search, file search, and image analysisRunning code/scripts and GitHub/API usageinit command and CLAUDE memory filePRD generation with business context + writing stylesCustom slash commands (stored prompts)Plan Mode vs auto-executionSub-agents and parallelizationMCP tool integrations (e.g., Reddit)PM use cases: research, docs, prototyping, automationsHybrid stack: Claude Pro + Cursor modelsInstagram growth + LLM meme system (Meme Mage)

In this episode of Aakash Gupta, featuring Aakash Gupta and Carl Vellotti, I should be charging $999 for this Claude Code Tutorial explores claude Code tutorial for PMs: workflows, agents, PRDs, automation wins Claude Code’s key advantage is moving beyond chatbot prompting into a file-system-first workflow where it can read, write, and refactor real project artifacts (docs, transcripts, code) with minimal friction.

At a glance

WHAT IT’S REALLY ABOUT

Claude Code tutorial for PMs: workflows, agents, PRDs, automation wins

  1. Claude Code’s key advantage is moving beyond chatbot prompting into a file-system-first workflow where it can read, write, and refactor real project artifacts (docs, transcripts, code) with minimal friction.
  2. The episode demonstrates practical PM workflows—summarizing customer interviews, generating PRDs with business context and style guides, and converting meeting transcripts into structured action items—using folders as “context engineering.”
  3. Claude Code differentiates from IDE-first tools (Cursor/Copilot/Lovable/Replit) by being more flexible for non-coding work, stronger at writing, and highly capable in the terminal with reliable tool use and task planning.
  4. Advanced capabilities include Plan Mode (planning without editing), custom slash commands (saved prompts), sub-agents for parallel work and role-based reviews, and MCPs to add new tools like Reddit access.
  5. The conversation closes with why Claude Code grew quickly (strong coding focus, writing quality, intent understanding) and a detour into Carl’s Instagram growth system, including an LLM-assisted meme creation tool (“Meme Mage”).

IDEAS WORTH REMEMBERING

7 ideas

Claude Code’s real unlock is file-system-native context, not “better prompts.”

By working inside a project folder, Claude can discover, read, and modify relevant artifacts (interviews, meeting notes, specs, code) without constant copy/paste, making “context engineering” fast and repeatable.

Run `init` early to create a durable project memory (CLAUDE file).

The CLAUDE file becomes always-on guidance that Claude references every session (repo structure, setup instructions, rules like “never commit without asking”), reducing repeated setup and preventing common mistakes.

Use Plan Mode to avoid “manifestation hell” on complex tasks.

Plan Mode (Shift+Tab) prevents file edits while Claude proposes steps and outputs, letting you correct structure/format before execution—especially important when generating many files or running multi-step coding tasks.

Turn recurring PM workflows into slash commands for consistent structure.

Custom commands act like stored prompts (e.g., meeting notes format with action items/metrics/risks) so outputs match your team’s expectations and voice, reducing the “did you use ChatGPT?” vibe.

Sub-agents enable two different accelerators: parallelization and perspective diversity.

Claude can spawn multiple agents to analyze multiple inputs at once (three interviews in parallel) or to critique the same artifact from distinct roles (designer/engineer/executive), improving speed and coverage.

MCPs (tool connectors) expand Claude Code from “smart editor” to “tool-using worker.”

Adding MCPs (e.g., Reddit) lets Claude reliably fetch/parse sources that are otherwise hard to access, enabling pipelines like ongoing competitor/user pain-point monitoring that feed directly into docs or prototypes.

A practical PM stack can be Claude Pro for docs + Cursor for heavier coding models.

The guest recommends using Claude Pro/Sonnet for writing/research workflows, and leaning on Cursor’s model access for more demanding coding, balancing cost with capability.

WORDS WORTH SAVING

5 quotes

It takes you out of the interface of just a chatbot, and it lets you build new workflows.

Carl Vellotti

As we’ve moved from prompt engineering into… context engineering, what you can give the LLM to work with is so key.

Carl Vellotti

The last thing that makes this really cool… is it can actually run code.

Carl Vellotti

There’s a little dopamine burst as you see it work through [the checklist].

Carl Vellotti

Have LLMs do the work that you hate, not the work that you love.

Carl Vellotti

QUESTIONS ANSWERED IN THIS EPISODE

5 questions

Exactly what should you put in the CLAUDE file on day one for a PM (rules, formats, repo map), and what should you avoid adding?

Claude Code’s key advantage is moving beyond chatbot prompting into a file-system-first workflow where it can read, write, and refactor real project artifacts (docs, transcripts, code) with minimal friction.

In what scenarios does “more context” hurt output quality in Claude Code, and how would you structure folders to minimize irrelevant tokens?

The episode demonstrates practical PM workflows—summarizing customer interviews, generating PRDs with business context and style guides, and converting meeting transcripts into structured action items—using folders as “context engineering.”

Can you show a concrete example of a slash command library a PM team might share (meeting notes, PRD, competitor brief, exec update), including file naming conventions?

Claude Code differentiates from IDE-first tools (Cursor/Copilot/Lovable/Replit) by being more flexible for non-coding work, stronger at writing, and highly capable in the terminal with reliable tool use and task planning.

How would you design a repeatable eval harness in Claude Code for testing prompts/models (like your transcript summarizer demo) without leaking API keys or producing noisy comparisons?

Advanced capabilities include Plan Mode (planning without editing), custom slash commands (saved prompts), sub-agents for parallel work and role-based reviews, and MCPs to add new tools like Reddit access.

Where do Claude Code sub-agents meaningfully beat a single-agent workflow, and where do they mostly waste tokens?

The conversation closes with why Claude Code grew quickly (strong coding focus, writing quality, intent understanding) and a detour into Carl’s Instagram growth system, including an LLM-assisted meme creation tool (“Meme Mage”).

EVERY SPOKEN WORD

Install uListen for AI-powered chat & search across the full episode — Get Full Transcript

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