What is vibe coding?
A working style where you steer a coding agent with intent and taste — and the agent writes the syntax, runs it, and shows you what broke. The craft moves from typing code to specifying outcomes and judging output.
What it actually is.
Intent over syntax
You describe what you want; the model writes the syntax. The skill is precision in language, not memorizing APIs.
Loops, not one-shots
Generate → run → read errors → refine. Speed comes from tightening the loop, not from typing faster.
Taste filters output
The model will happily ship garbage. Vibe coders develop taste — the gut sense that something is wrong before tests catch it.
Tools over models
What model you use matters less than what tools it can call. Files, browsers, runtimes, eval harnesses.
What it isn't.
Vibe coding is for non-engineers.
No. Senior engineers go fastest because they spot bad output earlier and steer harder. It compounds skill, it doesn't replace it.
It's just "using ChatGPT for code".
ChatGPT is one model in a chat box. Vibe coding is the practice: scoped tasks, agent tools, fast loops, runtime feedback, and disciplined evals.
It produces unmaintainable code.
It produces whatever you accept. The discipline is reading every diff and rejecting the ones you wouldn't have written yourself.
The Vibe Coder track.
12 weeks, 4 hours/week. Real projects with Claude Code, Cursor, and Aider. Capstone is shipping something to production.
See the track →