How I Use Vibe Coding on Every Project
Vibe coding is the practice of building software with AI coding assistants. Andrej Karpathy coined the term in early 2025, and by now it describes how a growing number of professional developers work every day. I am one of them. On every DevSac project, from a static Astro marketing site to a full-stack web app, I pair-program with Claude Code and Cursor. This page explains what that practice looks like in the room when I am building your project. I break down what vibe coding means for professional development in a dedicated article.
You are not hiring me to "use vibe coding" the way you hire someone to build a WordPress site. You are hiring me to build a WordPress site, or a web app, or an AI automation. Vibe coding is the method I use to deliver it. I document it here so you know what to expect when I show up with an AI in my toolbelt.
Faster Does Not Mean Lower Quality
The concern with AI-generated code is quality. That concern is valid when there is no experienced developer reviewing the output. AI coding assistants produce code that compiles and appears to work, but they make subtle mistakes: incorrect error handling, missing edge cases, security oversights, and architectural decisions that create maintenance problems later.
Every line of AI-generated code in my projects goes through the same review process as hand-written code. This is the core of AI pair programming: the AI accelerates the writing, the developer ensures the quality. You get both speed and reliability.
How I Use Claude Code and Cursor
Claude Code for web development handles complex, multi-file operations. Refactoring a component that touches 15 files. Adding a new API endpoint with validation, database queries, and error handling. Generating test suites for existing code. Claude Code reads the entire codebase, understands the patterns, and generates implementations that match the existing conventions.
Cursor handles real-time editing. Inline code suggestions while typing. Quick function implementations from a comment description. Rapid iteration on UI components with instant previews. The two tools complement each other: Claude Code for the heavy lifting, Cursor for the fine-grained work.
These AI tools work best when they have access to project-specific data. I build custom MCP servers as part of larger engagements to feed documentation, API schemas, database structures, and deployment configurations into the AI's context. The more the tool knows about your project, the better the generated code matches your requirements.
What This Means for Timelines and Cost
AI-assisted development reduces implementation time by 30-50% on well-scoped projects. A feature that takes 40 hours traditionally might take 25 hours with vibe coding. Architecture decisions, complex debugging, and integration work still take the same time. The savings come from faster boilerplate, instant pattern implementation, and rapid prototyping.
I pass that efficiency to clients through shorter timelines and lower project estimates. An MVP that would take 12 weeks ships in 8. A web application feature sprint that would take 4 weeks ships in 3. Same quality, less calendar time, lower total cost.
The Developer Still Matters
AI coding assistants do not replace the need for an experienced developer. They amplify the developer's productivity. A junior developer using Claude Code produces junior-quality code faster. A senior developer using Claude Code produces senior-quality code faster. The AI does not make architectural decisions, evaluate tradeoffs, or understand your business context. It generates code. The developer makes it right.
I have been building web applications, APIs, and SaaS platforms with TypeScript for years. AI coding tools make me faster at implementing what I already know how to build. That is the value proposition: senior engineering judgment at AI-assisted speed.