Skip to content
explainroo
Working with AI agents

What Is Vibe Coding?

Updated 3 min read

With explainroo How to Prompt a Coding Agent to Make an Explainer Video
On this page
  1. How vibe coding works
  2. What it is good for, and where it falls short
  3. How to use it with fewer surprises

If you describe in everyday language what you want software to do, an AI coding tool generates or changes the code. This approach is often called vibe coding. You guide the tool with follow-up prompts.

The term usually refers to a fast, conversational style of development. In its narrower sense, it also suggests accepting generated code with little review. That can work for a personal experiment or prototype, but it carries more risk when the software handles real users or important data.

How vibe coding works

You describe a goal, such as “Build a reading tracker where I can add books and mark them finished.” An AI tool generates code for the interface and other requested features. You run the result, point out what looks wrong, and ask for changes.

This back-and-forth means you spend less time writing code syntax and more time describing how the software should behave. For example, you might ask the tool to add a search box, change the layout, or explain why a button does not work. Tools used for this style include coding assistants built into editors and prompt-driven services such as Cursor, Replit, Lovable, and Bolt, as described in a beginner’s guide to vibe coding.

The term has a specific origin: computer scientist Andrej Karpathy coined “vibe coding” in February 2025. He described the approach as giving in to the “vibes” and, in effect, forgetting that the code exists. The phrase has since broadened to cover many kinds of prompt-driven development, including work where developers do inspect and test the generated code. That broader use makes it helpful to ask what someone means by vibe coding: AI-assisted coding in general, or coding with minimal review? The term’s history traces that shift.

What it is good for, and where it falls short

Vibe coding makes it easier to turn an idea into something people can try. A person without much programming experience can create a rough prototype, while an experienced developer can use prompts to generate boilerplate or explore a design. That speed is useful for demos, interface experiments, and small tools with limited consequences.

A working demo does not prove that the software works as intended. For instance, a reading tracker might display a list of books that was written directly into the page rather than saved to a database. The interface looks functional, but entries disappear or fail to update as soon as the user expects persistent storage. To avoid this, specify where data should be stored, how it should change, and who can access it.

Generated code can look correct while hiding flaws that a quick demo will not reveal. A 2026 audit summarized by CybeDefend examined 200 publicly deployed vibe-coded applications and reported at least one vulnerability in 91% of them. The audit found 1,186 issues, with 65.8% rated critical or high severity. Those figures describe the applications in that audit, not every project built with AI. The reported problems included broken access controls, injection flaws, and authentication failures.

This matters most when software handles payments, personal information, passwords, or other sensitive data. In those cases, a developer needs to understand what the generated code does, test security and behavior, and review changes before release. As developer Simon Willison warned, production software needs understandable code because engineers must safely change it over time. His comments on the risks highlight why a quick prototype and a maintained production service call for different levels of scrutiny.

How to use it with fewer surprises

Before asking the coding tool to build, give it specific requirements. Explain what the software does and what users can do, then describe how it should look and store and protect data. Make changes in small steps: ask for one feature, run it, and check the result before moving on.

Keep a record of code changes with version control, a tool that lets you review and restore earlier versions. Test the actual behavior rather than relying on a polished screen. With a reading tracker, add a book and refresh the page to confirm the entry remains. Then try invalid input and check whether one user can see another user’s data.

Use human review for anything that matters beyond a demonstration. The person releasing the software should be able to explain its important behavior, recognize when a test fails, and decide whether the code is safe to ship. If no one can do that, keep the project away from sensitive data and production use until someone with the necessary skills reviews it.

Free and open source

Let your AI agent make the video

explainroo lets a coding agent like Claude Code or Codex make narrated explainer videos and product demos. The voice, the word timing and the rendering run on your own computer, with no API key and no cost per video. Optional outside services use your own key.

Copy this into your AI agent

Make me a short explainer video about [your topic]. Use explainroo for it: run npx --yes explainroo docs and follow what it prints.

Getting started Example videos GitHub

Made with explainroo: How explainroo makes a video

More on working with ai agents

All guides on working with ai agents