CourseThe App Factory MindsetWhat Vibecoding Actually Is

What Vibecoding Actually Is

AI as engineer, you as PM. CLAUDE.md as onboarding docs3 min read

What Vibecoding Actually Is

Vibecoding isn't "having AI write code for you." That's like saying management is "having employees do work for you." It misses the entire point.

Vibecoding is a development methodology where you operate as the product manager and architect, and AI operates as the engineering team. You define what to build and why. AI handles how.

The Role Split

In traditional development:

  • You research → you design → you code → you debug → you deploy

In vibecoding:

  • You research → you design → AI codes → you + AI debug → you deploy

The critical insight: the most valuable work in software development is deciding what to build. That's what you keep. The implementation — translating decisions into syntax — is what AI handles.

CLAUDE.md: Your Onboarding Document

Every company writes onboarding docs for new engineers. CLAUDE.md is the onboarding document for your AI engineer.

It tells Claude (or any AI tool):

  • What the project is and why it exists
  • The architecture and key patterns
  • Coding conventions and style rules
  • What NOT to do (equally important)
  • How to build, test, and deploy
CLAUDE.md
# Project: BeautyTracker
 
## What This Is
iOS app for tracking skincare routines. Part of the Beauty Suite.
 
## Architecture
- SwiftUI + @Observable MVVM
- SwiftData for persistence
- AuraBeautyKit for shared components
 
## Conventions
- One file per type (View, ViewModel, Model)
- Feature-first directory structure
- No force unwrapping. Ever.
- Use SwiftData @Query, not manual fetching
 
## Build & Run
- `xcodegen generate` to regenerate project
- Build target: BeautyTracker (iOS 17+)
🔑Key Takeaway

The quality of your CLAUDE.md directly determines the quality of AI-generated code. A vague CLAUDE.md produces vague code. A specific, opinionated CLAUDE.md produces code that matches your standards.

The Three Modes of Vibecoding

1. Scaffolding Mode

You're starting a new project or feature. AI generates the initial structure.

Your job: Define the architecture, file structure, and patterns. AI's job: Generate boilerplate, create files, set up the skeleton.

2. Feature Mode

You're adding functionality to an existing project.

Your job: Describe the feature, its edge cases, and how it integrates. AI's job: Implement it following established patterns.

3. Fix Mode

Something's broken. You need to debug and repair.

Your job: Provide error messages, context, and constraints. AI's job: Diagnose and fix, following the build-fix loop.

What Vibecoding is NOT

Let's be clear about what this methodology isn't:

  • Not copy-pasting ChatGPT output — That's cargo cult programming
  • Not "no-code" — You need to understand code to review it
  • Not faster-but-worse — Done right, vibecodeed apps are indistinguishable from hand-coded ones
  • Not a shortcut around understanding — You need to understand architecture, patterns, and business logic. You just don't need to memorize syntax

The Compound Advantage

Vibecoding gets better over time because:

  1. Your CLAUDE.md gets more detailed — Each project teaches you what to specify
  2. Your PATTERNS.md grows — Solutions you've validated become reusable templates
  3. Your frameworks mature — Shared code means AI has better context
  4. Your prompting improves — You learn how to describe what you want precisely

This is why the portfolio strategy and vibecoding are synergistic. Each app makes the methodology stronger for the next one.

💡Tip

Start a PATTERNS.md file today. Every time you solve a non-obvious problem — a SwiftData migration pattern, a StoreKit edge case, a UI trick — write it down. This file becomes your institutional knowledge base.