Vibe coding meaning: What it is, who coined it and the tools to build apps by prompting

Plans, prices and features checked on the official pages on October 28, 2026.

Vibe coding means building software by describing what you want to an AI in plain language, then running the result and asking for changes, without reading or writing the code line by line. The term was coined by AI researcher Andrej Karpathy in a post on X on February 2, 2025. Today it covers a family of tools, from browser app builders to coding agents in the terminal.

This guide explains the meaning, how the process works, the main tools with the plan details published on their official pages, and where the approach goes wrong.

Vibe coding meaning: Describing an app instead of writing it

In the original post, as widely quoted, Karpathy described a way of working where you “fully give in to the vibes” and forget that the code exists: you ask the model for a change, accept it, run the program, and paste errors back until it works. The phrase stuck because it named something many people were already doing with chat assistants and AI editors.

It is called vibe coding because the person steers by feel and by results, not by understanding each line. If you review, test and understand the code the AI writes, many developers would call that AI-assisted programming instead. The line between the two is blurry, and the word is now used loosely for any app built mostly by prompting.

The idea is not tied to one product. A person with no programming background can use it to build a small tool, and an experienced developer can use it to produce a first version quickly.

How vibe coding works, step by step

Most tools follow the same loop, whatever their interface.

1. You describe the app in a prompt: what it does, who uses it, what the screens look like.
2. The AI generates the files and shows a live preview or runs the program.
3. You try it and reply with changes, bug reports or screenshots.
4. The AI edits the code and the preview updates.
5. When it works, you publish it, often with one click, or export the code to GitHub.

Good prompts are specific about behavior (“a form that saves entries and shows them in a table sorted by date”) and small in scope. Building one feature at a time makes errors easier to find. The shift from single prompts to AI that carries out multi-step tasks is covered in From prompts to management: Steering the new era of AI agents.

The main vibe coding tools and what they cost

The tools fall into two groups. App builders run in the browser and host what they make. Coding agents and AI editors work on code on your own machine or repository and suit people who can read code.

App builders in the browser

Replit builds and hosts apps in one workspace. Its pricing page lists Core at $20 per month ($18 per month billed annually), and describes Free Mode as capped at 30 hours of chat and 60 projects, with $20 of credit toward more powerful models. Pro is $100 per month ($90 billed annually) with 10 parallel agents, up to 15 collaborators and database rollback up to 28 days. Enterprise is custom. Replit also publishes tutorials for beginners.

Lovable generates full-stack web apps from a chat. According to its documentation, the Free plan gives 5 build credits per day, up to 30 per month, plus 20 monthly Cloud credits. Paid plans are Pro and Business, and Enterprise is custom; the pages consulted did not state monthly prices for Pro and Business. Plan Mode costs 1 credit per message, and Lovable says credits are non-refundable.

Bolt offers a free tier of 1M tokens per month with a 300K daily limit and Bolt branding on published sites. Pro is $25 per month with 10M monthly tokens with rollover, no daily limit and custom domains. Teams is $30 per member per month, and Enterprise is custom.

v0 from Vercel is aimed at web interfaces and deploys to Vercel. The Free plan allows 7 messages per day. Plus is listed at $30 per user per month with $30 in monthly credits and $2 in daily credits, and Business at $100 per user per month. New users get $5 in one-time credits.

Google AI Studio Build mode is described by Google as a vibe coding environment where you describe an app in natural language and an agent generates the code with a live preview. It can produce React and Node.js web apps or native Android apps, connect to GitHub and deploy to Cloud Run. Google’s page does not say whether Build mode itself is free; it notes that paid models and Cloud Run usage may incur costs. Google’s earlier no-code experiment is covered in Google’s Opal and Gemini.

Coding agents and AI editors

Claude Code from Anthropic is an agent that works in your codebase. Anthropic’s pricing page says it is included in all paid Claude plans: Pro at $20 per month billed monthly ($17 per month billed annually), Max from $100 per month, and Team and Enterprise seats. The Free plan does not include it. A real-world example of what developers do with it is in Developers are using Claude Code to strip ads and tracking from smart TVs.

Cursor is an AI code editor. Its pricing page currently lists tiers from a free plan to $200 per month, with a usage pool shared across Cursor and Grok, and says third-party models in Cursor are available at API list price. Check the page for the current tier contents before subscribing, as the structure has changed.

Mistral Vibe is the new name of Mistral’s assistant; its models and plans are in Mistral Le Chat is now Vibe.

Prices checked on October 28, 2026.

Which vibe coding tool is best for your goal

No tool wins on every task. Pick by what you want to end up with.

• A web app you can share, no coding background: an app builder such as Replit, Lovable or Bolt, which host the result.
• A polished web interface: v0, which is built around Vercel deployment.
• An existing project or serious software: a coding agent or editor such as Claude Code or Cursor, where you can read and version the code.
• An app that uses Gemini features: Google AI Studio Build.

Start with a free tier, build one small app and watch how fast the credits drop. Credit and token systems differ, and the cost of a task depends on its size, which is why the free limits above matter more than the headline price.

Where to start and where to learn

The simplest start is to choose one builder, write a one-paragraph description of a small tool you actually need, and iterate. A habit tracker, a price calculator or a simple booking form are realistic first projects. Official documentation and tutorials from the makers, such as Replit’s beginner material and the Lovable docs, are the most reliable place to learn how each tool handles credits and publishing.

If you want to go further, learn the basics of what the AI produces: what a database is, what an API key is, and how to read an error message. You do not need to write code, but this knowledge lets you catch mistakes.

Why vibe coding goes wrong and how to limit the risk

The common criticism is that code nobody reads is code nobody can trust. Typical problems are bugs that appear when you add features, security gaps such as exposed keys or missing access checks, and apps that become hard to change because the AI keeps rewriting parts that worked.

Practical safeguards:

• Keep secrets such as API keys in the tool’s settings, never in a prompt or in public code.
• Do not collect personal data or take payments with an app you have not had reviewed.
• Use version control so you can go back to a working version.
• Ask the AI to explain what it changed and to write tests for important logic.

The security side of AI misuse, from the other direction, is covered in Anthropic vs AI cybercrime.

Common questions