Vibe Coding
Vibe coding is a way of building software in which a person describes what they want in plain language, an AI model writes the code, and the person judges the result mainly by running it, not by reading or fully understanding the code. AI researcher Andrej Karpathy coined the term in February 2025, describing it as giving in to the vibes and forgetting that the code even exists.
How Vibe Coding Works
A vibe coding session is a fast conversational loop. The builder asks for a feature, the AI generates or edits code, and the builder runs the app to see what happens. If something breaks, the usual move is to paste the error message back to the AI and ask for a fix. If the fix does not work, the builder asks for a different approach or works around the problem.
The defining feature is what the builder skips. In Karpathy's original description, the person accepts changes without reviewing the diffs and lets the code grow beyond their own understanding. The developer and writer Simon Willison draws the line clearly: if you review, test and can explain the AI-written code, you are doing software development with AI help, not vibe coding.
Vibe coding usually happens in AI-enabled editors, chat tools or an AI coding agent that can run the project and edit many files at once.
Why Vibe Coding Matters
Vibe coding has made software creation accessible to people who could not build before, and it has made prototyping cheap for people who could. A founder can test an idea with a working demo in an afternoon instead of writing a brief and waiting weeks.
The trade-off is control. Code nobody has read can hide security flaws, duplicated logic and fragile shortcuts, which builds technical debt quickly. Product decisions also get lost: each prompt carries only part of the intent, so behavior drifts from what was originally agreed. Willison notes that vibe coding your way to a production codebase is clearly risky, because most real engineering work means evolving systems whose code must stay understandable.
The practical takeaway is to use vibe coding where speed matters more than durability, such as throwaway prototypes and personal tools, and to add structure before code reaches real customers.
Vibe Coding Example
A solo founder wants to test whether gym owners would use a class waitlist tool. She opens an AI app builder and types: "Build a page where members join a waitlist for a full class and get an email when a spot opens." She clicks through the result, asks for a cleaner layout, pastes two error messages back to the AI, and has a clickable demo by evening. She never opens the database code.
That is a good use of vibe coding: the goal is learning, not production. When three gym owners ask to pilot it, she writes down the rules (who gets the spot first, how long they have to claim it) and moves to a reviewed, specified build. Keeping those decisions consistent as the product grows is covered in vibe coding at scale across epics and sprints.
Vibe Coding vs. Vibe Coder
Vibe coding is the practice. A vibe coder is the person: a builder, often without formal engineering or product training, who ships mainly through AI tools. A vibe coder can choose to work in a more structured way, for example with spec-driven development, and many do once a prototype becomes a real product.