What Is Vibe Coding

Vibe Coding: Writing Code by Describing the Vibe

Vibe coding is the practice of writing software by describing what you want in natural language to an AI, then accepting the output without deeply reviewing or understanding every line. You describe the "vibe" of what you need — "make a dashboard that shows user analytics with filters" — and the AI generates the code.

The term was coined by Andrej Karpathy in a tweet in February 2025. He described it as: "There's a new kind of coding I call 'vibe coding', where you fully give in to the vibes, embrace exponentials, and forget that the code even exists." He was half-joking, but the label stuck because it captured something real about how many developers were already using AI tools.

In vibe coding, you're not writing functions or debugging logic — you're having a conversation with an AI and directing it toward the outcome you want. The code is a byproduct of that conversation.

The Pros of Vibe Coding

• Speed — You can build a working prototype in hours instead of days. The AI handles boilerplate, setup, and routine logic while you focus on what the product should do.
• Lower barrier to entry — Developers who are less experienced in a particular framework or language can still produce working code. You don't need to know the React hooks API by heart to build a React app.
• Exploration — Vibe coding is great for testing ideas. You can rapidly spin up three different approaches to a problem and pick the one that works best.
• Fun — There's a genuine creative satisfaction in describing what you want and watching it come to life. It feels like having a very fast, very patient pair programmer.

The Cons of Vibe Coding

• You don't understand your own code — When you accept output without reviewing it, you end up with a codebase you can't debug, modify, or explain. This becomes a real problem the moment something breaks.
• Hidden bugs and security issues — AI-generated code can look correct but contain subtle bugs, insecure defaults, or logic errors that only surface in production.
• Technical debt accumulates fast — Vibe-coded projects tend to have inconsistent patterns, duplicated logic, and poor architecture because each AI response is generated in isolation.
• Hard to maintain — When you need to add a feature three months later and don't understand the existing code, you're forced to either vibe code more changes on top (making it worse) or rewrite from scratch.

When Vibe Coding Works (And When It Doesn't)

Vibe coding works well for:

• Prototypes and MVPs — When you need to validate an idea quickly, vibe coding lets you ship something functional in a day.
• Internal tools — Scripts, dashboards, and utilities used by a small team where bugs are tolerable and the stakes are low.
• Learning and exploration — Building something to understand a concept, not to put into production.
• Boilerplate generation — Getting the skeleton of a project set up quickly, then filling in the critical logic yourself.

Vibe coding fails for:

• Production systems — Anything that handles user data, processes payments, or needs to be reliable.
• Security-critical code — Authentication, authorization, encryption. These require careful review that vibe coding explicitly skips.
• Performance-sensitive applications — AI tends to generate straightforward but not optimized code. If you need sub-millisecond responses, you need to understand every line.
• Long-lived codebases — Code you'll maintain for years needs to be understood by humans, not just generated by AI.

How Vibe Coding Affects Costs

Vibe coding has a peculiar cost profile. On the surface, it seems cheap:

• Fewer turns per session — Vibe coders tend to give broad instructions and accept the output, rather than iterating on specific details. A vibe-coded feature might take 5 turns vs. 15 turns for a carefully reviewed one.
• Zero-shot is cheaper — If you get the output you want in one prompt, you only pay for one round of tokens. No context buildup, no compounding costs.

But the hidden cost is rework. When a vibe-coded feature breaks and you need to fix it, you're often starting from scratch because no one understands the code well enough to make targeted fixes. That means new sessions, more turns, and more tokens — often more than you would have spent building it carefully the first time.

The most cost-effective approach? Vibe code the first draft, then review and understand what was generated. You get the speed of vibe coding with the maintainability of intentional development. The review pass costs a few extra turns, but it saves you from expensive rework later.