Vibe Coding
Vibe coding is an AI-assisted programming approach where a developer describes desired behavior in natural language and an AI model generates the corresponding code, with the human focusing on intent rather than implementation details.
Vibe coding is a development style in which a programmer delegates most or all code-writing to a large language model, interacting through conversational natural-language prompts rather than writing syntax manually. The term was coined by Andrej Karpathy in a February 2025 post on X and quickly spread through the developer community as a label for a workflow many practitioners had already begun adopting.
The typical workflow involves describing a desired feature or fix in plain English inside an AI-assisted editor — tools such as Cursor, Windsurf, or GitHub Copilot Workspace — then accepting, testing, or iterating on the generated output. The developer acts primarily as a specification writer and quality arbiter rather than a line-by-line coder; error messages are often pasted directly back into the chat for the model to self-correct. Version control and diff review replace manual code reading as the primary quality checkpoint.
Vibe coding lowers the barrier to building functional software, enabling non-programmers to create working prototypes and allowing experienced engineers to explore ideas at dramatically higher speed. It shifts engineering judgment toward architecture, test coverage, and security review rather than syntactic fluency, and raises practical concerns about code ownership, long-term maintainability, and silent bugs introduced by accepted-but-unread AI output.
By mid-2026, frontier models such as Claude Sonnet 4, GPT-4o, and Gemini 2.5 Pro sustain multi-file, multi-session vibe coding workflows with moderate human supervision. Dedicated environments with persistent agent contexts, integrated terminal access, and inline diff views have become standard. Enterprise adoption has grown alongside code-review gates and static analysis tooling designed to catch correctness and vulnerability issues before AI-generated code reaches production.