Vibe coding is not about writing less code. It is about thinking at a higher level of abstraction...
What Vibe Coding Actually Is
Vibe coding is a development methodology where you describe what you want to build in natural language and collaborate with AI to produce working software. You set the direction, make architectural decisions, and review outputs. The AI writes the implementation.
This is not "no-code." You need to understand code to review it, debug it, and direct the AI when it goes off track. But you spend your time on architecture and design decisions rather than syntax and boilerplate.
The Toolchain
Primary: Claude Code
Claude Code is the foundation of my vibe coding setup. It operates in the terminal with full filesystem access, which means it can read your project structure, understand your codebase, and make changes across multiple files in a single operation.
The key advantage over chat-based AI coding is context. Claude Code sees your entire project — types, configurations, existing patterns — and generates code that fits.
Supporting Tools
- Cursor — For visual editing when I need to see the full file context
- GitHub Copilot — For inline completions during manual editing
- Supabase — For backend services that would take days to build from scratch
- Vercel — For deployment that requires zero configuration
The Process
Phase 1: Architecture First
Before writing any code, define the architecture in plain language. Write a document that describes:
- What the app does
- What the data model looks like
- What the user flows are
- What the tech stack is and why
This document becomes the AI's brief. The better this document, the better every subsequent output.
Phase 2: Scaffold and Iterate
Start with the hardest part. Not the login page — the core feature that makes the app worth building. Give Claude Code the architecture document and ask it to build the data model and core API first.
Review every file. Not line by line, but structurally. Does the data model match your architecture? Are the API patterns consistent? Are the types correct?
Phase 3: Build Outward
Once the core works, build outward: UI components, secondary features, integrations. Each layer references the patterns established in the core. Consistency compounds — by the third feature, the AI is generating code that matches your project's conventions without being told.
Phase 4: Polish and Ship
The last 20% takes 80% of the time, even with AI. Edge cases, error handling, loading states, mobile responsiveness, accessibility. This is where human judgment matters most because these details define the difference between a prototype and a product.
The Mindset Shift
From Writer to Editor
Traditional development: you write code. Vibe coding: you review code. The skill shifts from "can I implement this?" to "is this implementation correct?" This is a fundamentally different cognitive mode — and it is faster once you adjust.
From Perfectionist to Pragmatist
AI-generated code is not always how you would write it. That is fine. The question is not "is this my style?" but "does this work, is it maintainable, and does it meet the requirements?" If yes, ship it.
From Sequential to Parallel
You can direct AI to build multiple features simultaneously. While one implementation is being reviewed, the next is being generated. Your throughput multiplies when you stop thinking sequentially.
When Vibe Coding Breaks Down
It is not a silver bullet. Vibe coding struggles with:
- Novel algorithms — AI is great at applying known patterns, less so at inventing new ones
- Complex state management — Multi-step state machines need human design
- Performance optimization — AI optimizes for correctness, not always for speed
- Domain-specific edge cases — If it is not in the training data, you are on your own
Know when to vibe and when to write. The best developers do both.
Getting Started
- Pick a project small enough to complete in a week
- Write the architecture document before opening your editor
- Use Claude Code to scaffold the core feature
- Review, iterate, and ship
The first project will feel slow as you calibrate. The second will feel fast. By the third, you will not want to go back.
Explore Frameworks
More from Building in Public
Building Beauty Bar: Week 2
Auth, payments, and the first real user test. Week 2 of building a salon booking app live with AI.
Why I Build in Public
The case for radical transparency in the AI age. Why sharing failures accelerates learning faster than hoarding wins.
Building Beauty Bar: Week 1 of Vibe Coding a Full App
I'm building an AI-powered booking app for salons. Here's what happened in week 1 — the good, the broken, and the agents that saved it.
The AI Alchemist
Practical AI strategies, behind-the-scenes builds, and emerging tools — delivered weekly to practitioners.