What Is Vibe Coding? A Beginner’s Guide With a Practical Example

What Is Vibe Coding? cover showing hands holding hand-drawn interface and chat sketches next to a laptop.

Vibe coding is building software by describing what you want to an AI, trying the result, and asking for changes in plain language. In the original sense, you rely on what you see the software do and spend little time reading the code it writes. People also use the phrase more broadly for prompt-driven app development.

A simple example is a kitchen measurement converter. You ask for a page that converts grams to kilograms, notice that a blank input returns zero, and describe what should happen instead. The example shows why the approach helps you test an idea quickly and where code review becomes useful.

Where did the term come from?

Andrej Karpathy introduced the phrase in a February 2025 post about building software with a language model while paying little attention to the generated code. The original usage helps distinguish an experiment driven by the visible result from a development workflow where you review and maintain every change. Karpathy’s original post.

Developer Simon Willison describes vibe coding as building with a language model without reviewing its code. The term can therefore describe one phase of a project. A prototype can start that way, then move into a workflow where someone inspects changes, adds tests, and takes responsibility for maintenance. Willison’s explanation of the distinction.

Headline of Simon Willison’s article explaining how vibe coding differs from AI-assisted programming.
Simon Willison’s article explains the narrower meaning of vibe coding.

A worked example with a kitchen converter

The goal is one page with a number field, a direction selector, and an answer. Keep it local and use made-up inputs. Leaving out accounts, storage, and external services keeps the first prototype small enough to understand and easy to replace.

Build a local web page that converts grams to kilograms and kilograms to grams. Show the units next to the input and result. Accept zero and decimal amounts. For a blank or negative value, show a helpful message. Keep the feature on one page without network requests. List three example inputs and expected results.

Open the page and test both directions. For example, 750 grams should return 0.75 kilograms, and 1.25 kilograms should return 1,250 grams. Check that zero is accepted. Leave the field blank and see whether the page asks for an amount. Changing the direction should also update the unit labels.

If leaving the field blank returns zero, ask for a narrow change. “When the amount field is blank, show ‘Enter an amount’ and clear the previous answer. Keep zero valid.” Retest the example cases plus the new one. This helps you see when a fix accidentally changes behavior that already worked. For more on setting constraints and specifying outcomes, read Unlimited AI’s prompt-writing guide.

The roles of prompting, testing, and reviewing

ActivityWhat you learnConverter example
Describe the behaviorWhether the AI has enough information to build what you want.State both directions and explain how blank input should behave.
Use the prototypeWhether the visible result works for the cases you try.Check 750 grams, zero, a decimal, and a blank input.
Review the changesHow it handles inputs and which files or dependencies it adds.Read the conversion and validation logic and check that the page makes no network requests.
Add repeatable testsWhether the same behavior survives future edits.Save the expected outputs and test them after changes.

A successful demo tells you the app worked in that moment. A review can answer questions the screen can’t. For instance, the request said to keep the page local, so you should check whether the generated files load anything over the network before handing them to someone else.

Before other people depend on the result

  1. Write down what the prototype promises. Include the normal path, empty input, invalid input, and anything the program stores or sends.
  2. Review the files and changes. Ask a knowledgeable reviewer to check what changed and why. GitHub documents a file-by-file review process that lets a team track and discuss individual changes.
  3. Test the real environment. Repeat the important cases on the devices and browsers your audience uses. Include keyboard navigation if it’s a web page.
  4. Decide who will maintain it. Keep the source, basic setup instructions, and expected behavior together so someone can reproduce the app and repair it.

If the project starts handling accounts, payments, or personal information, the cost of a mistake changes. Get the right review before using real data or letting people rely on the result. GitHub’s guide to reviewing proposed changes.

GitHub pull request navigation with the Files changed tab highlighted.
GitHub’s Files changed tab lets reviewers inspect the proposed edits file by file.

Common questions about vibe coding

Do you need to know how to code?

You can describe and try a simple prototype without writing the first version yourself. Reading basic code helps when behavior is wrong or you need to understand what will happen when other people use it.

Is all AI-assisted coding vibe coding?

In the original usage, vibe coding relies heavily on the visible result with little inspection of the generated code. AI-assisted coding covers a wider range of workflows, including ones where a developer reviews the implementation and runs tests.

What is a good first vibe coding project?

Pick a small task with an answer you can verify, such as a unit converter or a mock event schedule made with sample data. That gives you a quick way to see whether each requested change works.

Leave a Comment

Your email address will not be published. Required fields are marked *


Scroll to Top