Vibe Coding

Jun 19, 2025 | Updated Jun 19, 2025

Vibe coding is a novel software development approach that involves using Artificial Intelligence to generate code.

What Is Vibe Coding?

Vibe coding (AI vibecoding) refers to a new approach to software development that allows users to create executable code with simple natural language commands. In other words, you give prompts to an artificial intelligence (AI) model, and the model generates software code.

Andrej Karpathy, OpenAI co-founder, coined this phrase in February 2025 to define a form of writing code using AI assistance. This shifts the programmer’s task from the conventional approach of manually writing code to simply guiding, testing, and improving the AI-generated code. As such, no deep programming knowledge or training is necessary to be a vibe coder.

How Does It Work?

Technically speaking, this hands-off coding style employs complex machine learning algorithms to translate natural language instructions into code. As such, it takes a more conversational approach, where you simply issue voice commands in plain language while the AI tool generates the corresponding source code.

You simply accept all the suggestions the AI gives without fully understanding the code. You also refrain from tweaking the generated code. Instead, you just “go with the vibes.” That said, vibe coders place their trust in the AI model. When faced with an error, they simply copy and paste it into the tool to fix it.

In short, vibe coding can be characterized by:

Machine learning – It uses machine learning models trained on extensive programming data to generate code based on user input.
Natural language processing (NLP) – It relies on NLP models to comprehend and interpret user input from natural language into a software language.
Real-time feedback – You receive and provide real-time feedback to the coding tool, facilitating effective iteration and improvement of the generated code.
Accessibility – With AI vibecoding, even non-tech-savvy individuals can develop functional applications without extensive programming knowledge. In other words, it eliminates the entry barriers to software development.

While it is easy and more accessible than conventional coding, it is not suited for large-scale, complex projects. Instead, it can be used for low-stakes, throwaway, or hobby projects. This is because AI-generated source code comes with security flaws, buggy code, and legal ownership issues. In addition, the AI-generated output harbors the risk of insufficient reviews and sign-off.