Vibe Coding
FoundationsA prompt-driven development style in which an AI writes much of the implementation and the developer steers the result through natural language, tests, and review.
It is improvisational theatre with a compiler backstage: the human sets the scene, the model fills in the moves, and tests decide whether the performance holds.
Vibe coding is the playful name Andrej Karpathy gave to a development workflow in which the programmer lets an AI assistant produce most of the code and guides it through conversation. The human describes intent, reviews the result, reports failures, and keeps nudging the implementation toward a working outcome.
The workflow changes the division of labor. The model becomes an implementation engine for scaffolding, refactors, and glue code; the developer supplies goals, constraints, architecture, taste, and acceptance tests. Tools such as Cursor, GitHub Copilot, Claude Code, and similar environments make the loop practical by combining chat, editing, execution, and debugging.
Vibe coding is fast because it removes much of the typing bottleneck, but generated code still needs evidence. Without tests and review, the process can hide security flaws, incorrect assumptions, and maintenance debt behind a convincing demo. It is well suited to prototypes and exploration; production work needs a clear owner for the design and the final code.
Related Terms
Last updated: February 25, 2026