Aakash GuptaI should be charging $999 for this Claude Code Tutorial
CHAPTERS
- 0:00 – 1:46
Claude Code’s breakout growth + what you’ll learn in this tutorial
Aakash frames Claude Code’s explosive adoption and positions the episode as a beginner-to-hero tutorial. Carl is introduced as both a senior PM and the creator behind the largest PM Instagram account, setting up why his workflows are worth copying.
- •Claude Code’s rapid ARR growth and tool displacement claim
- •Episode promise: become a “Claude Code hero” and build a personal copilot
- •Why Carl’s perspective matters (PM + heavy Claude Code user + agent builder)
- 1:46 – 2:42
Why non-technical PMs should care: moving from chatbot prompts to workflows
Carl explains that Claude Code’s biggest unlock is leaving the chat UI and working directly in a file system. This enables repeatable workflows, reusable prompts, and faster context engineering using real project artifacts.
- •Limitations of chat interfaces for ongoing PM work
- •Claude Code as a flexible workflow layer over your files
- •Context engineering as the core advantage (vs one-off prompt engineering)
- 2:42 – 5:10
Claude Code vs Cursor/Copilot/Lovable/Replit + why terminal CLIs matter
The conversation compares Claude Code to IDE-first assistants and other CLIs. Carl argues Claude excels at writing and intent, while Claude Code’s terminal-first interface is unexpectedly powerful and polished compared to newer CLI entrants.
- •IDE tools are code-centric; Claude Code is broader (research, docs, PRDs)
- •Claude’s writing quality vs ChatGPT’s recognizable “voice”
- •CLI competitors (Gemini CLI, OpenAI Codex) and why Claude Code feels most polished
- •Pricing note: Claude Pro can run Claude Code (not necessarily $200/mo)
- 5:10 – 7:38
Install and first run: one-command setup + basic commands
Carl walks through installation from Anthropic’s quickstart and launches Claude Code in Terminal. He shows the chat-like interaction model, introduces context management with “clear,” and emphasizes you can talk naturally without learning many commands.
- •Native install via single terminal command
- •Launching with `claude` and chatting in plain language
- •Using `clear` to reset and manage context
- 7:38 – 11:48
Working in a project folder: querying your files like a lightweight knowledge base
Carl switches from the computer root to a specific demo folder that resembles a company wiki. Claude Code can enumerate files, read interviews, and produce comparative analysis across documents without manual copy/paste.
- •Open Claude Code inside the folder you want it to understand
- •Ask file-aware questions (count interviews, summarize one, compare segments)
- •Demonstrates fast, practical context engineering for PM artifacts
- 11:48 – 14:31
Web search + token visibility: research directly from the terminal
Claude Code performs a live web lookup (example: new iPhone) and reports findings. Carl highlights the UI benefit of seeing token usage, which makes cost/context tradeoffs more concrete than standard chat products.
- •Built-in web search workflow from Claude Code
- •Token counter reveals how quickly context is consumed
- •Prompt quality matters (example search returns partial mismatch)
- 14:31 – 15:24
Beyond text: bringing images into Claude Code for analysis
Even in a text-first terminal interface, Carl demonstrates dragging an image in and receiving feedback. He notes this is useful for reviewing graphics or debugging via screenshots.
- •Attach images directly to Claude Code sessions
- •Use cases: feedback on visuals, debugging via screenshots
- •Still benefits from context hygiene (clearing when needed)
- 15:24 – 18:03
Running code with Claude Code: GitHub repo + generating a YouTube transcript file
Carl shows the step-change from chatbots: Claude Code can execute code to produce real outputs. Given a transcript API repo and a YouTube link, Claude creates a task plan, runs the script, and writes a markdown file with results.
- •Claude Code can run scripts and write files locally
- •Auto-generated to-do plans keep work structured and sequential
- •Non-technical users can leverage GitHub repos via natural language
- 18:03 – 21:14
Using Claude Code inside an IDE (Cursor) + cost/model strategy
They open the same workflow in Cursor’s integrated terminal while intentionally not using Cursor’s sidebar assistant. Carl explains a hybrid approach: use Claude Pro/Sonnet for docs and workflows, and Cursor’s model access for heavier coding when needed.
- •Run Claude Code anywhere there’s a terminal (including Cursor)
- •Hybrid pricing/ROI discussion for PMs (Claude Pro + Cursor)
- •Preview and edit generated markdown outputs inside the IDE
- 21:14 – 24:07
The `init` command and the CLAUDE file: persistent project memory + rules
Carl introduces `init`, which generates/updates a CLAUDE file describing repo structure and operating instructions. He explains how CLAUDE acts as durable “always-on” guidance (rules, formats, writing preferences) compared to fragile chat context.
- •`init` builds a project overview and instructions in CLAUDE
- •CLAUDE functions like persistent memory across sessions
- •Add rules quickly with a `#` entry (e.g., ask before Git commits)
- •Scoped CLAUDE files can exist in subfolders for localized rules
- 24:07 – 32:59
PRD generation with context engineering: business info + style guides + examples
Using a structured folder of business context, writing styles, and an example PRD, Carl runs a “super prompt” to draft a technical PRD for a GPT real-time feature. Claude researches unknown details, follows the style, and outputs a formatted draft with links.
- •Organize reusable context: business info, styles, exemplar PRDs
- •Claude does web research + internal file reading in one workflow
- •Outputs a strong PRD first draft (problem, goals, constraints, references)
- •Discussion of context overload and why clearing/one-shot context helps
- 32:59 – 34:29
Operational PM workflows: auto-summarizing meeting transcripts + action items
Carl demonstrates using Claude Code to modify existing documents by appending summaries and action items. This highlights the advantage over chatbots: Claude can directly update files in place and standardize outputs at scale.
- •Process a folder of transcripts and append summaries/action items
- •File-editing capability as a differentiator vs chat
- •Action items can include owners and due dates even with minimal prompting
- 34:29 – 39:20
Custom slash commands: saved prompts for consistent notes, tone, and follow-ups
Carl introduces slash commands as reusable stored prompts, enabling consistent formatting for meeting notes. He also shows combining a tone/style file with meeting context to draft a personalized Slack follow-up that doesn’t feel like generic AI writing.
- •Slash commands = stored prompts triggered on demand
- •Standardize outputs (metrics, risks, next steps) across meetings
- •Blend context + writing style to avoid “Did you use ChatGPT?” vibes
- •Brief Slack integration note (via MCP, often send-to-self then copy)
- 39:20 – 45:21
Plan Mode + multi-session productivity: safer execution and parallel work
Carl toggles Plan Mode (Shift+Tab) to force planning before edits, avoiding wasted runs. He then runs multiple Claude instances concurrently, showing how you can keep working while long tasks execute in the background.
- •Plan Mode prevents edits and forces a checklist plan first
- •Example: building a transcript summarizer eval workflow (prompts + model comparisons)
- •Refine plan structure before execution (Markdown outputs, file organization)
- •Run separate Claude sessions in parallel for throughput
- 45:21 – 56:40
Agents and subagents: parallel analysis, role-based reviews, and agent libraries
Carl demonstrates subagents to parallelize work across multiple interviews and to get different perspectives (designer/engineer/exec). He shows how to define agents as simple config/text files, import new agents (e.g., legal) from public registries, and why agent personalities change outputs.
- •Parallelize tasks by spawning subagents per document
- •Role-based review: designer/engineer/executive perspectives into a file
- •Agent definitions as lightweight files (colors/personas)
- •Importable libraries (e.g., subagents.cc) and extending with org-specific context
- 56:40 – 1:03:31
MCPs (tooling extensions): adding Reddit and other external systems to Claude Code
The episode moves from “vanilla” Claude Code to tool augmentation through MCPs. Carl demos a Reddit MCP to extract pain points from threads—useful for research pipelines—and discusses registries for MCPs like Google Drive and other integrations.
- •MCPs extend Claude Code with LLM-friendly tool APIs
- •Reddit MCP enables reliable extraction/analysis from threads
- •Use cases: recurring content/research digests, competitor monitoring
- •Ecosystem note: many tools skew engineering today; PM-focused agents/tools are an opportunity
- 1:03:31 – 1:11:24
Prototyping a workflow builder + practical guidance (evals, front-end tweaks, safety)
Carl shows Claude Code generating a simple workflow-builder prototype from a spec: node creation and connections on a canvas. They discuss using Claude for eval workflows, when front-end vibe coding is appropriate, and the importance of Plan Mode, supervision, and safe environments.
- •Prototype demo: canvas/grid + add nodes + connect nodes
- •Claude Code for evals: generating prompts, running tests, integrating into workflow
- •Front-end tweaks can be feasible; production readiness still needs engineering review
- •Avoid “manifestation hell” via Plan Mode + tighter requirements + human oversight
- 1:11:24 – 1:24:32
Why Claude Code took off + best PM use cases + Claude vs automation platforms
They step back to explain the product’s momentum: strong coding focus, superior writing, and intent understanding. Carl summarizes top PM use cases (research, writing-heavy workflows, lightweight prototyping) and contrasts Claude Code’s tactical agents with recurring automation tools like n8n/Lindy.
- •Adoption drivers: developer focus, writing quality, intent understanding
- •Best PM use cases: research, PRDs/notes/docs, simple prototypes
- •Claude Code agents (tactical) vs n8n/Lindy automations (recurring workflows)
- •When to use single vs multi-agent systems (complexity vs reliability tradeoff)
- 1:24:32 – 1:29:23
Carl’s Instagram growth playbook: memes, shareability, and consistency
The conversation shifts to Carl’s creator journey and how he built the largest PM Instagram account. He explains platform dynamics (Reels), what formats spread (PM–engineering jokes, “PM not doing work”), and how consistent posting plus DMs compounds growth.
- •Started on Twitter; moved memes to Instagram/Threads dynamics
- •Instagram virality favors shareability over wit
- •High-performing meme themes: PM–engineering relationship, PM-as-non-worker trope
- •Cadence: ~2 posts per weekday for ~2.5 years + story reposts + DM replies
- 1:29:23 – 1:37:04
AI for content creation: Meme Mage and Carl’s next chapter (newsletter/community)
Carl reveals Meme Mage, his internal tool using video understanding (Gemini) plus meme templates and personas to generate caption candidates. He closes with his transition from PM to building The Fullstack PM newsletter/community, currently focused on growth over monetization.
- •Meme Mage: templates + personas + video analysis to draft meme captions
- •LLMs provide “90% there” options; human punch-up still required
- •Left his PM job to build The Fullstack PM
- •Current focus: community/newsletter growth (no monetization yet)
