Short answer
Vibe coding is building software by describing what you want in plain language to an AI model and letting it write, run and fix the code, with the human guiding by results ("make the button bigger", "it crashes when I log in") rather than reading every line. The term was coined by AI researcher Andrej Karpathy in February 2025 and was named Collins Dictionary's Word of the Year for 2025. It is great for prototypes and personal tools, but risky for production software unless someone reviews the code.
Where the term came from
On February 2, 2025, Andrej Karpathy, a co-founder of OpenAI and former head of AI at Tesla, posted about a new way of programming where you "fully give in to the vibes" and forget the code even exists. He described accepting AI changes without reading the diffs, pasting error messages back to the AI and letting it fix them.
The phrase spread quickly because it captured something real: large language models had become good enough that non-programmers could build working apps, and professional developers could build far faster. By November 2025, Collins Dictionary had picked "vibe coding" as its Word of the Year.
How vibe coding works
1. Describe the goal: "Build a web page that shows today's GPU cloud prices in a sortable table." 2. Let the AI build it: an AI coding tool writes the files, installs packages and runs the app. 3. Test by using it: you click around and see what works. 4. Give feedback in plain words: "The prices should be in dollars per hour" or paste the error message. 5. Repeat until it does what you want.
Modern tools work as AI agents: they can read the whole project, run commands and tests, browse documentation and fix their own mistakes in a loop. Many connect to services like GitHub or databases through MCP.
Popular vibe coding tools
| Tool type | Examples | Best for |
|---|---|---|
| AI app builders (browser-based) | Lovable, Bolt, Replit, v0 | Non-programmers building web apps and prototypes |
| AI code editors | Cursor, Windsurf, VS Code with GitHub Copilot | Developers working in real codebases |
| Terminal and cloud coding agents | Claude Code, OpenAI Codex, Gemini CLI | Multi-step tasks across a whole repository |
| Chat assistants | ChatGPT, Claude, Gemini | Small scripts, explanations and one-off tools |
Vibe coding vs AI-assisted programming
Not all AI-assisted coding is vibe coding. Developer Simon Willison and others draw a useful line:
| Vibe coding | AI-assisted engineering | |
|---|---|---|
| Reads the code? | Rarely | Yes, reviews every change |
| Understands how it works? | Not necessarily | Yes |
| Tests | "It seems to work" | Automated tests and review |
| Best for | Prototypes, personal tools, learning, demos | Production systems, team codebases |
| Responsibility | Often unclear | The developer owns the code |
Professional teams increasingly use AI for most of the typing while engineers still review, test and take responsibility for what ships.
What vibe coding is good for
- Prototypes and MVPs: test an idea in hours instead of weeks.
- Personal tools: a budget tracker, a file renamer, a dashboard for your own data.
- Internal tools: simple admin panels or reports that only a few people use.
- Learning: seeing working code for an idea, then asking the AI to explain it.
- Websites and landing pages: fast first versions that a designer or developer can refine.
- Data analysis: quick scripts to clean a spreadsheet or draw a chart.
The risks of vibe coding
- Security holes: AI-generated code can include injection flaws, exposed API keys, missing authentication or weak database permissions. Several vibe-coded apps have leaked user data because nobody checked access rules.
- Hidden bugs: code that works in a demo can fail with real users, edge cases or larger data.
- Maintainability: a large codebase nobody understands becomes hard to change; the AI may also struggle once the project grows beyond what fits in its context window.
- Hallucinated libraries and APIs: models sometimes call packages or functions that do not exist, and attackers have registered fake package names to exploit this. See AI hallucinations.
- Costs: agents running many steps consume a lot of tokens, and cloud services the app uses can generate surprise bills.
- Licensing and compliance: you are still responsible for the code you publish.
How to vibe code more safely
- Keep secrets (API keys, passwords) out of code and out of chats; use environment variables.
- Ask the AI to add tests, then run them.
- Ask a second pass to review the code for security issues, or have a developer review anything that handles user data or payments.
- Use version control (Git) so you can undo bad changes.
- Start small, commit often, and keep the project structure simple.
- Turn on authentication and database access rules before sharing an app publicly.
Does vibe coding need special hardware?
No. Most vibe coding tools run on cloud models, so any laptop works. If you want privacy or to avoid usage costs, you can run open coding models locally, but good coding models need plenty of VRAM: a 24 to 32 GB GPU or a high-memory Mac handles mid-sized models well. See how to run an LLM locally and the best GPU for local LLMs.
Frequently asked questions
Can someone with no coding experience vibe code?
Yes. Browser-based app builders let beginners create working web apps from descriptions. Understanding basic concepts (what a database, API and login are) helps a lot when something goes wrong.
Will vibe coding replace programmers?
It is changing the job more than removing it. AI writes more of the code, while engineers spend more time on design, review, testing, security and deciding what to build. Complex, reliable systems still need people who understand them.
Is vibe coding good for learning to code?
It can be, if you ask the AI to explain what it wrote and you experiment with changes. Accepting code blindly teaches little.
Which AI model is best for vibe coding?
The leading coding models change often. Frontier models from Anthropic, OpenAI and Google are the most capable today, and reasoning modes help with harder bugs. See What is a reasoning model?.
Sources and further reading
- Euronews, Vibe coding named word of the year by Collins Dictionary (November 2025)
- Simon Willison, Not all AI-assisted programming is vibe coding (March 2025)
- Wikipedia, Vibe coding




Leave a Reply