> glossary / vibe-coding
Vibe coding
A programming workflow where a developer prompts an AI coding agent in natural language and reviews the intent of the output rather than the syntax.
Coined by Andrej Karpathy in a February 2025 social-media post, "vibe coding" describes a style of software development in which an AI agent with tool-use capabilities (Claude Code, Cursor, Gemini CLI, Codex CLI) writes and edits files on the developer's behalf while the developer reviews outcomes — running the program, checking tests, watching the dev server — rather than reading every diff line-by-line.
Vibe coding is distinct from three adjacent workflows. AI autocomplete (tab-to-accept in an editor) keeps the human in the typing loop. LLM pair programming (chat with ChatGPT, paste into editor) keeps the human in the edit-boundary loop. Agentic refactors where every diff is read still count as traditional engineering. Vibe coding drops both of these guards.
The workflow has strong critics — code-provenance concerns, security drift, skill atrophy — and strong proponents, especially among solo builders and greenfield work. VibeMon is built for this workflow: the unit of measurement is the agent's hook event, not the human's keystroke.
Related terms
- Drop — The unit of feeding in VibeMon. One drop is awarded per eligible hook event from a supported AI coding agent.
- Hook event — A structured message that an AI coding agent emits at a defined point in its lifecycle. VibeMon's hook endpoint receives these and converts them into drops.