Coined by Andrej Karpathy, vibe coding describes a mode of development where you lean into the AI — describing what you want rather than writing how to do it — and largely trust the model’s output. You steer by intent, not by line.

It’s not a single thing. It’s a spectrum:

What changes at the high end of the spectrum:

• The bottleneck shifts from writing code to specifying intent clearly
• Token budget management becomes a real concern — long sessions lose context
• Code review becomes the primary skill, not code writing
• Domain knowledge and systems thinking matter more than syntax fluency

The honest tradeoff:

Vibe coding accelerates exploration and prototyping dramatically. The risk is accumulating code you don’t understand — which surfaces as bugs you can’t debug and architecture that drifts.

The skill is knowing when to go deep and when to trust the output.

See also:

• Tokens as AI Currency
• Developer Role in AI-Assisted Development
• Barefoot Developer Philosophy