What Is Vibe Coding?
Vibe coding is building software by describing what you want to an AI coding tool and iterating on what it generates instead of writing every line yourself; the term was coined by Andrej Karpathy in February 2025, and VibeKing lists 2,620 apps built this way.
At a glance: vibe coding vs AI-assisted coding, the tools, 12 real apps, liveness data
| AI-assisted coding | Vibe coding | |
|---|---|---|
| Who writes the code | The developer, with AI completions and suggestions | The AI, from natural-language prompts |
| Who reads the code | The developer reviews every change | Often nobody; the human reads the running result, not the diff |
| The human's job | Design, write, review, test | Describe, run, react, re-prompt |
| Typical tools | Copilot completions, inline chat in an IDE | Agents and app builders: Cursor agent, Claude Code, Replit, Lovable, Bolt, v0 |
| Best suited to | Systems that must be maintained for years | Prototypes, personal tools, small products, experiments, and a growing number of real launches |
| Main trade-off | Slower, but the code is understood | Fast, but bugs and security issues can ship unread |
| Tool | What it is | Apps tagged on VibeKing |
|---|---|---|
| Cursor | AI-native code editor with chat and agent modes that edit whole projects | 4 → |
| Claude and Claude Code | Anthropic's model, used in chat or as a terminal coding agent | 9 → |
| ChatGPT and Codex | OpenAI's chat model and its coding agent | 2 |
| GitHub Copilot | Completions, chat and agent mode inside VS Code and GitHub | 1 |
| Replit | Browser IDE with an agent that builds, runs and hosts full apps | 1 |
| Lovable | Prompt-to-web-app builder with hosting and a Supabase back end | 2 |
| Bolt | Prompt-to-app builder from StackBlitz that runs in the browser | none tagged yet |
| v0 | Vercel's generative UI tool for React and Next.js front ends | none tagged yet |
| Gemini | Google's model, in Gemini chat, AI Studio and the Gemini CLI | 1 |
| GPT-4 class models | Used directly through the API or inside custom agents | 3 → |
| Windsurf | AI editor with the Cascade agent | none tagged yet |
| # | App | What it does | Category | Built with | Upvotes | Status | Link |
|---|---|---|---|---|---|---|---|
| 1 | VibeSail | Multiplayer sailing sim — $8K MRR | Game | GPT-4 + Copilot | 92 | Online | vibesail.com ↗ |
| 2 | WebsiteToPrompt | Clone any website with AI prompts | Dev Tools | Claude | 88 | Online | websitetoprompt.com ↗ |
| 3 | ChatIQ | AI customer support — $2K MRR | SaaS | Claude + GPT-4 | 61 | Online | chatiq.ai ↗ |
| 4 | Savvit | Should you buy now or wait? | Lifestyle | Cursor + Claude | 60 | Online | savvit.app ↗ |
| 5 | Seedance 2.0 Mini | Create lightweight cinematic AI videos from text and images | Video | Seedance 2.0 Mini AI | 59 | Online | seedancemini.net ↗ |
| 6 | Seedance 2.0 | Create consistent AI videos from images, audio, and text | Video | Seedance AI | 57 | Online | www.seedancepro.net ↗ |
| 7 | Qconcursos | Ed-tech rebuilt by 2 devs instead of 30 | Education | Lovable | 54 | Online | qconcursos.com ↗ |
| 8 | Scail 2 | Animate characters from reference images without complex rigging | Video | Scail 2 AI | 53 | Online | scail2.com ↗ |
| 9 | PodGen | AI podcast generator — full SaaS | SaaS | Claude 4 | 51 | Online | podgen.io ↗ |
| 10 | Disko | SMS messaging service — $11K MRR | SaaS | Replit | 44 | Online | disko.is ↗ |
| 11 | Plaintext CRM | CRM that runs in your IDE | SaaS | Claude Code | 44 | Online | github.com ↗ |
| 12 | LaunchBert | Your product deserves to be found. Let's launch it! | Marketing | PHP, Laravel, Livewire, Alpine.js… | 43 | Online | launchbert.com ↗ |
The 2,620 listings span 15 categories; the largest are AI Coding (691), Dev Tools (405), Marketing (320). Browse them all on the category index or start with the overall best-of list.
How much vibe-coded software stays online?
As of 2026-09-11, VibeKing had re-probed 2,620 of its 2,620 listed apps at their public URL: 2,397 responded normally (91%), 48 were intermittently unreachable, and 175 were offline. Liveness is not quality, but it is a hard number on how much vibe-coded software stays up after launch.
Vibe coding, defined
Vibe coding is building software by describing what you want to an AI coding tool in natural language, running the code it generates, and steering it with further prompts instead of writing and reviewing every line by hand. The developer's attention moves from the code to the result: you run the app, see what is wrong, and describe the fix. Errors get pasted back into the tool rather than debugged by hand.
The phrase comes from a post by Andrej Karpathy on X in February 2025:
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.Andrej Karpathy, X, February 2025
The term spread fast enough that Collins English Dictionary chose vibe coding as its Word of the Year for 2025. On VibeKing, "vibe-coded" is used in the broad sense: an app whose maker built it primarily by prompting AI coding tools.
Vibe coding vs AI-assisted coding
Both use AI, so the two get conflated. The useful distinction is who reads the code: in AI-assisted coding the developer still reviews every change, while in vibe coding the human reads the running result, not the diff. The side-by-side comparison is in the At a glance panel above.
The tools people vibe code with
Any large language model can vibe code through a chat window, but the practice took off with tools that run, edit and deploy code on your behalf. These are the tools VibeKing tracks; the At a glance panel above lists each one with how many listings state it.
12 real vibe-coded apps
The At a glance panel above lists 12 live VibeKing listings with what each does, its category, the build tool when the maker states it, upvotes and liveness status. Click through to any app page for the maker, the listed date and a structured facts table.
Where vibe coding falls short
- Unread code is unreviewed code. Security mistakes such as exposed keys, missing auth checks and injection bugs ship easily when nobody reads the diff. Anything that handles money or personal data needs a human review pass.
- Debugging by re-prompting has a ceiling. When the model cannot fix a bug from the error message alone, the developer has to understand the code after all, and by then the codebase can be large and unfamiliar.
- Maintenance is the hidden cost. Generated code tends toward duplication and inconsistent structure, which is fine for a weekend project and painful for a product with paying users.
- Prompts are not specs. The tool builds what you described, not what you meant. Small, testable steps beat one giant prompt.
How to start vibe coding
- Pick a tool that can run code, not just write it: an agentic editor such as Cursor or Claude Code, or a hosted builder such as Replit, Lovable or Bolt.
- Describe one small feature at a time, run it, and paste errors straight back into the tool.
- Commit to version control at every working checkpoint so a bad prompt is a revert, not a rewrite.
- Before anyone else uses it, read the parts that handle auth, payments and user data, or have a reviewer do it.
- Ship it, then list it on VibeKing with the tool you used so the built-with pages stay honest.