Vibe coding

Andrej Karpathy, a founding member of OpenAI and former head of AI at Tesla, named the practice in a post on X on 2 February 2025. He described "a new kind of coding" where you "fully give in to the vibes, embrace exponentials, and forget that the code even exists." In his telling you talk to the model, accept its changes, paste error messages back in, and judge the result by running it, not by reading the diff.

The phrase spread fast. Merriam-Webster listed it as slang and trending in March 2025, and Collins Dictionary named "vibe coding" its Word of the Year for 2025.

The defining feature is that you don't review the code. Simon Willison put the line plainly: vibe coding means "building software with an LLM without reviewing the code it writes." If a model wrote every line but you read, tested and understood it, that's ordinary software development with a fast typist. The distinction matters because the two carry different risks. A reviewed change is as safe as your review. An unreviewed one is as safe as the model's defaults and whatever checks run automatically.

It works well for prototypes, internal tools, personal projects and first versions, where speed matters more than polish and the person prompting is also the user. It breaks in predictable places once real users arrive:

• Security. A database that returns every row looks the same as one that returns your rows while you're the only user. Leaked keys and permissive access policies are the classic failures.
• Maintainability. Code nobody has read is hard to change safely. Each new prompt works from the model's understanding of the code, not yours.
• Debugging loops. When the model has a wrong picture of the code, more prompts tend to reinforce it rather than fix it.

The practical answer isn't to stop vibe coding. It's to put automatic checks where human review would have been: type-checking, builds, security scans and a written plan before large changes.

Vibely is an AI app builder built for this way of working. You describe a web app or a native iOS and Android app, and the agent writes React and TypeScript code (or an Expo app for mobile) with a Supabase backend. The checks a reviewer would have done run automatically: every turn that changes files is type-checked and built before Vibely finishes, a basic security scan starts on every publish, and Plan mode lets you approve a written plan before a large change touches the code. Because the output is real code you can read, sync to GitHub and download, you can stop vibe coding and start reviewing whenever the project calls for it.