Vibe coding is the process of using prompts to get AI to generate code. There are both pros and cons to vibe coding.
Vibe coding was coined in 2025 by Andrej Karpathy, one of the co-founders of OpenAI. It’s even been added to the Merriam-Webster dictionary slang section.
How does vibe coding work?
Traditional coding is the process of writing code in different programming languages. Vibe coding, on the other hand, works by using prompts to generate code with Large Language Models (LLM).
Vibe coding is often done in stages. For example, you could say “Generate code for a booking form where people can book my services.” Once the outline of the form has been generated, you could say “Add a dropdown for this list of services.” And so on.
Once the LLM has generated the code, further prompts are used to make changes to the code. This means that you don’t have to be a developer or know any coding language to create an app, a website, or write any kind of code.
The pros of vibe coding
As mentioned, one of the main pros of vibe coding is that it can open up the world of coding to more people. If you, for example, need to create something simple for your website, prompting an LLM to generate the code is much easier than learning to code yourself.
The LLM can also help you test what it’s made, and you can discuss your project with the LLM and get some good suggestions for ways to solve problems.
However, there are some drawbacks to vibe coding too.
The drawbacks of vibe coding
Since you don’t know code language, you can’t go through and verify the code that the LLM has generated. In fact, this is one of the defining features of vibe coding: no review and no editing of the actual code.
This means that you have to test what’s been generated, and then make prompts to correct any errors you find. That makes vibe coding an iterative process. You might struggle to get rid of all bugs and issues, since new iterations of the code can fix one problem and cause another. The resulting code might not be very consistent, and it could be difficult for an actual developer to look at it later, or edit and maintain it.
LLMs also have a tendency to hallucinate, which can cause problems and inconsistencies in the code. On a smaller scale, LLMs can forget details as you keep coming back to a project, and so you’ll have to spend time on retraining it and giving it context for what it’s supposed to be doing.
You also need to make sure that you take data security into account. It’s a good idea to get a real developer to test and verify that there are no potential data leaks or risks in your projects.
All this to say, that vibe coding can be a great tool to allow more people to generate code, but it’s important to have checks in place to verify the output from the LLM and in some cases, you might have to take the time to learn something yourself or at least ask a developer you know!
