AI Agent Glossary · AI Coding Tools Glossary
Vibe coding
Building software by describing what you want to an AI and accepting its output without reading the code, relying on behaviour rather than inspection.
Andrej Karpathy coined the term in a February 2025 post: you "fully give in to the vibes, embrace exponentials, and forget that the code even exists". You describe what you want, run what the AI produced, paste any error back, and keep going without reading the code.
What it is not. Using an AI to write code is not automatically vibe coding. Simon Willison drew the line in March 2025: if an LLM wrote the code and you then reviewed it, tested it and could explain how it works, "that's not vibe coding, it's software development". The defining trait is that nobody inspects the output.
Example. Someone asks an AI for a script that renames a folder of holiday photos by date, runs it on a copy, sees the names look right, and never opens the file. That is vibe coding, and for a one-off job on their own data it is a reasonable trade.
Where it works. Prototypes, one-off scripts, internal tools with a single user, and learning what is possible before a real build.
Where it breaks. Anything that must be secured, maintained or extended by someone else. In Veracode's 2026 benchmark, 44% of AI code-generation tasks produced code with a security flaw. If nobody reads the code and no tests or scanners run, those flaws ship, and the next change starts from code no one understands.
How common it is. Y Combinator told TechCrunch that a quarter of its Winter 2025 batch had codebases about 95% AI-generated. Yet in Stack Overflow's 2025 survey, 72% of developers said vibe coding is not part of their professional work.