Series: Gstack (1/7)

Why AI Assistants Need 'Role Switching' — The Core Logic of Gstack →
Gstack Beginner Course: Turn Claude Code into Your Virtual Dev Team
#gstack#claude-code#AI dev workflow#AI virtual dev team#gstack tutorial#AI productivity#claude-code-skills

Gstack Beginner Course: Turn Claude Code into Your Virtual Dev Team

5 min read

About two weeks into using Claude Code to write software, you'll probably hit a wall: AI is obedient, but it never asks "are you sure you want to do this?"

You say add a feature, it adds it. You say refactor this logic, it refactors. You say fix this endpoint, it fixes it — then introduces two new bugs, and you spend three more conversation rounds cleaning up.

The problem isn't that Claude isn't smart enough. The problem is that you keep making it use the same brain for everything: product decisions, architecture design, code review, testing, deployment — all crammed into one context, one conversation to rule them all.

That's exactly what Gstack is designed to fix.

What This Series Solves

Gstack (full name GStack) is an open-source Claude Code skills framework built by Y Combinator CEO Garry Tan. The core concept is one sentence:

"Planning is not review. Review is not shipping. Founder taste is not engineering rigor. If you blur all of that together, you get a mediocre blend of all four."

Plain English: put AI in the right cognitive mode at the right time — don't let it run on one mode for everything.

Gstack provides nine skills, each corresponding to a different role in the software development cycle: product decisions with CEO thinking, architecture design from an engineer's perspective, paranoid senior engineer code review, automated QA testing, one-command deployment — behind every slash command is a pre-configured cognitive mode.

This series takes that framework from installation to real-world scenarios, explained clearly across six posts.

Who This Series Is For

  • People already using Claude Code but getting inconsistent output quality and having to iterate repeatedly
  • Engineers who want to reduce manual intervention in review, testing, and deployment
  • Solo developers building products who want to upgrade AI from a "typing tool" to a "virtual dev team"
  • Anyone who saw Garry Tan's 10,000 lines of code per week number and wants to know how he actually does it

If you haven't used Claude Code yet, this series may move too fast — get comfortable with Claude Code first, then come back.

Series Roadmap

These six posts start from "why you need Gstack," move through installation and detailed explanations of each core skill, and finish with an integrated demo using a real development scenario:

1. Why AI Assistants Need "Role Switching" — The Core Logic of Gstack Build the mental model first — Gstack doesn't solve the problem of AI not being smart enough, it solves "not giving AI the right cognitive mode." After reading this, you'll understand why nine skills beat one universal prompt.

2. Gstack Installation and Quick Start: Nine Skills in 10 Minutes From GitHub clone to completed setup, with solutions to common installation issues. You'll be able to run your first slash command after reading this.

3. Plan Review Dual Mode: CEO Thinking + Engineer Architecture — Why to Use Them Separately /plan-ceo-review and /plan-eng-review are where Gstack delivers the most value per use — one asks "should we build this," the other asks "how should we build it right." This post explains the difference and when to use each.

4. Code Review and Ship: From Paranoid Review to One-Command Delivery /review finds the problems that pass CI but explode in production. /ship compresses sync, test, push, and PR into one command. This post explains the internal logic of both and how to combine them.

5. QA and Browse: Give AI Eyes to Actually "See" Your App /qa analyzes your git diff, finds affected pages, tests them, fixes bugs, and verifies — outputting a health score from 0 to 100. /browse lets AI click through your app with a real browser. This post explains why they're worth setting up.

6. Real-World Dev Scenario: The Complete Gstack Workflow from New Feature to Production Connects the concepts from the previous five posts using a real scenario — starting from /plan-ceo-review and running all the way through to /ship.

1 → 2 → 3 → 4 → 5 → 6 in order is the most solid approach — build the concepts, then install, then use each skill, then run the full workflow.

Short on time: read at least 1 (understand why you need it) and 2 (actually get the tool installed). The rest you can read based on your needs, but 3 has the highest ROI — don't skip it.

Quick Start Version

One sentence: Install Gstack → run /plan-ceo-review + /plan-eng-review before new features → run /review when implementation is done → run /ship when everything looks good.

That's it. The six posts handle the details and edge cases. This post has one job: show you what the map looks like, so you can pick a path and start.