What is vibe coding?
Vibe coding is a way of developing where you don't read or review the code yourself. You give the AI plain-language instructions and keep going based only on whether the result works. The term was coined by Andrej Karpathy in February 2025.
The phrase captures the feeling that you are not "writing code" so much as describing the vibe you want and letting the AI build it.
How do you do vibe coding?
You describe what you want to build in plain language, run what the AI produces, and if it isn't what you wanted, ask for changes in words. You repeat this loop. An AI coding tool or agent does the actual implementation.
Its strength is building prototypes quickly, judging by "does it work?" instead of understanding each line of code.
How is vibe coding different from regular AI coding?
In typical agentic coding, the developer gets help from AI but still reads, reviews, and controls the code. Vibe coding mostly skips code review and judges only by how the result behaves, so the level of control is different.
That makes vibe coding a good fit for quick experiments and prototyping, while production services where verification matters call for a more careful approach.
The reality and risks of vibe coding
You can produce results fast, but if code piles up without being understood, it is easy to miss security vulnerabilities and hidden bugs, and maintenance can get hard.
So it is useful for prototypes and personal projects, but for important services you will eventually need to understand the code and verify it with tests.