What is vibe coding? A practical guide to building with AI
Kofi Agyem, Founder and CEO, CloudzyroPublished · 8 min read
On this page
Vibe coding means building software by describing it to an AI and running what it writes. Where the term came from, the tools, the risks and how to ship.
1. Vibe coding in one paragraph
Vibe coding is building software by describing what you want to an AI, running what it writes and asking for changes, without reading most of the code. You judge the result by whether it works, not by how the code looks. It's the fastest way yet to go from an idea to something you can click on, and it's brilliant for prototypes and small tools. It needs more care once real people, payments or personal data depend on it.
This guide covers where the term came from, what it's good and bad for, the tools, a workflow that works, the risks worth taking seriously and how to put what you build online.
2. Where the term came from
The AI researcher Andrej Karpathy coined the phrase in a post on X in February 2025: “There's a new kind of coding I call ‘vibe coding’, where you fully give in to the vibes, embrace exponentials, and forget that the code even exists.” He described it as “not too bad for throwaway weekend projects”, adding: “I just see stuff, say stuff, run stuff, and copy paste stuff, and it mostly works.”
The phrase spread fast. Collins Dictionary named “vibe coding” its Word of the Year for 2025, defining it as “the use of artificial intelligence prompted by natural language to write computer code”.
There's a useful distinction here, made by the developer Simon Willison: if you review, test and understand the code an AI writes, that's ordinary software development with an AI helper. Vibe coding, strictly, is when you don't. Most real projects sit somewhere between the two, and the closer you get to real users, the further toward the careful end you should move.
3. What vibe coding is good for
- Prototypes. Show an idea working in an afternoon instead of describing it in a document. See hosting for prototypes.
- MVPs. A first version to put in front of real customers and learn from. See hosting for MVPs.
- Small business sites. A landing page, a menu, a portfolio or a booking form.
- Internal tools. A dashboard over a spreadsheet, a form that saves to a database, a calculator your team uses every day.
- Learning. Asking an AI to build something, then asking it to explain each part, is a quick way to pick up how the web works.
4. Where it struggles
- Large, long-lived codebases. Without someone who understands the structure, each change can quietly break something else.
- Security-sensitive features. Sign-in, payments, file uploads and anything with personal data are easy to get almost right, which is the dangerous kind of wrong.
- Data you can't lose. AI tools sometimes save data to local files or make database changes you didn't ask for.
- Debugging loops. When a fix breaks something else and the next fix breaks the first thing, you need to stop and understand the problem rather than keep asking.
None of these rule vibe coding out. They mean you slow down, read the parts that matter and test them yourself.
5. The tools
Vibe coding tools fall into three groups. Which you choose depends on how much control you want.
- Chat assistants: Claude and ChatGPT. Describe a site and get the files back. Great for small static sites, and both can deploy for you once connected to a host.
- Coding agents: Claude Code, Codex, Cursor, Gemini CLI and Windsurf. They work in a folder on your computer, run commands, read errors and fix them. The best choice once a project has more than a handful of files. See hosting for coding agents.
- App builders: Lovable, Bolt, v0 and Replit. A browser-based editor with a live preview. The quickest way to start, especially if you've never opened a terminal. Lovable and Bolt usually make React apps with Vite, and v0 usually makes Next.js apps.
You don't have to pick one forever. A common path is to start in an app builder, sync the code to GitHub, then keep going with a coding agent once the project grows.
6. A realistic vibe coding workflow
- Write a short brief. Who it's for, the three or four things it must do, and what it should look like. A paragraph is enough. Paste it into every new conversation.
- Pick a boring stack. Ask for something common, such as plain HTML, Vite with React, Next.js or a small Node.js server. Popular stacks are what AI tools know best, and they're easy to host.
- Build one feature at a time. Ask for the smallest version that works, run it, then ask for the next piece. Big requests produce big, tangled changes.
- Use version control. Commit to Git after each working step, or ask your agent to. When a change goes wrong, you can go back instead of untangling it.
- Paste errors back in, with context. Include what you did, what you expected and the full error. If the same fix fails twice, ask the AI to explain the cause before it changes anything.
- Keep data in a database. Ask for a hosted database such as Supabase or Neon rather than files on disk. See how to connect Supabase.
- Review the risky parts. Before real users arrive, read the sign-in, payment and database code, or ask a fresh AI session to review it for security problems.
- Put it online early. A live link shows problems a local preview hides, like missing environment variables or the wrong port.
7. The risks, and how to handle them
Secrets in the code. AI tools happily paste API keys straight into files, and those files end up in Git, in zips and sometimes in the browser. Keep keys in environment variables instead. Remember that anything named with VITE_ or NEXT_PUBLIC_ is built into the JavaScript every visitor downloads, so only public values belong there. Our guide to environment variables explains the difference.
Open databases. Many vibe-coded apps talk to Supabase straight from the browser. That's fine only with Row Level Security turned on: Supabase's own docs warn that a table in an exposed schema without it “is readable and writable by any role with a grant on it”. Ask your AI tool to enable it on every table and write policies, then test them while signed out.
Data saved to local files. If an app saves uploads or a SQLite database next to its code, that data usually disappears on the next deploy on modern app hosts. Keep it in a database or a storage service.
Code nobody understands. Ask the AI to write a short README explaining how the app is put together, keep it updated, and ask for explanations of anything you'll need to change later. Future you, or a developer you hire, will thank you.
Dependencies. AI tools sometimes suggest packages that are out of date, or that don't exist at all. Check unfamiliar package names before you install them, and keep your framework updated.
8. How to put a vibe-coded app online
Most of what vibe coding tools make is either a static site (HTML, CSS and JavaScript, often built by Vite) or a Node.js app (Next.js, Express and similar). Both need a host that builds the code and serves it on an address with HTTPS. You have three ways to get it there:
- Ask your AI tool. Claude, Claude Code, ChatGPT, Cursor and Codex can deploy to Cloudzyro through our connector. Connect once, then say “Put this site online with Cloudzyro”. Your tool creates the site, sends the files and follows the build. It can never see your secrets, delete a site or change your plan. Step by step: deploy a site built with Claude.
- Connect GitHub. Lovable, Bolt, v0 and Replit can sync your code to a GitHub repository. Connect that repository to your site and every push goes live by itself. GitLab, Bitbucket and other Git addresses work too.
- Upload a zip. Download your project and upload the folder or zip in the console.
On Cloudzyro, the free Launch plan allows business use, gives each site 1 CPU, 512 MB of memory and 100 GB of traffic a month, and never puts sites to sleep. Your own domain is free on every plan, and every deploy is kept as a version you can restore in one click. Sites have no permanent disk, so data belongs in a database such as Supabase or Neon. See Sites & Apps for what's supported.
9. Should you vibe code?
If you have an idea and no developer, yes. It's the cheapest way there has ever been to find out whether something is worth building. Start small, put it online early, and treat the code as a draft until real users depend on it. When they do, spend an afternoon on the boring parts: secrets, database rules, backups and a read through the code that handles money and personal data.