What is vibe coding? A plain-English guide
Learn what is vibe coding in plain terms: how it works step by step, a real example, and the risks to check before you launch.
On this page
Vibe coding is a way of building software by describing what you want in plain language and letting an AI write the code, then refining the result through conversation. You don't need to know a programming language to start, but you do need to check whether the result actually does what you asked. This guide explains what vibe coding is, shows how it works with a concrete example, and covers its strengths and risks, so you can decide whether to try it on a small app.
What is vibe coding, in one sentence
Vibe coding means describing an app or feature in ordinary words, having an AI generate the code, and then steering the build with follow-up requests like "make the buttons bigger" or "only admins can see this page." The person sets the goal and judges the result. The AI does the typing.
Where the term came from
Andrej Karpathy, a well-known AI researcher, popularized the term in early 2025. The phrase caught on because it described something many non-programmers were already doing: they trusted the AI's output, ran it, and kept adjusting until it felt right, often without reading the code closely.
What vibe coding is not
Vibe coding is not the same as no-code. No-code tools give you fixed blocks, like pre-built form widgets and workflow templates, that you arrange on a canvas. Vibe coding produces real software: source code, a front end, a back end and a database that you can change, extend or hand to a developer later. The output is still software, which is why the risks in the later sections matter.
How vibe coding works, step by step
The workflow is the same whether you use a chat-based AI app builder or a coding assistant. It has four steps.
- Describe the goal in plain language. Say who the app is for, what it should do and what data it needs. Skip the technical vocabulary. "A page where customers pick a service and a time, and I see the bookings in a list" is a good first prompt.
- Review the first working version. The AI will produce something runnable, often within minutes. Click through it as a real user would. Note what is missing or wrong.
- Ask for changes in small steps. Request one change at a time: "add a phone number field," "send the list to a spreadsheet," "hide cancelled bookings."
- Test the result yourself before sharing it. Try the app with realistic data and edge cases: an empty form, a double booking, a wrong password. Don't rely on the AI to report that everything works.
A worked example: a booking page for a barber shop
Imagine a barber who wants customers to book appointments online without phone calls. Here is how a vibe-coded session might go.
- Prompt 1: "Build a booking page for a barber shop. Customers choose a haircut, a date and a time slot, and enter their name and phone number. I want to see all bookings on an admin page."
- Review: The barber opens the preview and sees a form and a list. The time slots are not limited to opening hours.
- Prompt 2: "Only offer slots between 9:00 and 18:00, Monday to Saturday, in 30-minute steps."
- Review: Customers can now book the same slot twice.
- Prompt 3: "Prevent two bookings in the same slot. Show a clear message when a slot is taken."
- Test: The barber books three appointments with the same time, changes a date, and tries an empty phone number. One fix is needed: the phone field accepts letters. He asks for a number-only check and retests.
Each round is small, the result is checked against a real task, and the barber never reads the code. That is the core rhythm of vibe coding. If you want a longer walkthrough of this type of project, see Build a booking app with AI.
Why small, specific requests work better
Vague prompts like "make it better" produce changes you can't evaluate. Specific requests give the AI a clear target and give you a clear test. A useful rule: if you can't say how you would check the change in under a minute, the request is too broad. Split it.
Vibe coding examples people actually build
Most successful vibe-coded projects are modest and practical. Common examples include:
- Internal tools: a lead list with status tags, a request form that sends submissions to a shared table, or an approval queue for small teams.
- Simple dashboards: a page that reads a spreadsheet and shows totals, charts or overdue items for the week.
- Landing pages and client portals: a page to collect sign-ups, or a login-protected page where a client downloads their files.
- Small booking and catalog apps: scheduling for a local service, or a product list for a small shop.
Each of these needs three things before you start. First, clear inputs: what data goes in, such as a spreadsheet, a form or a file. Second, a clear user: who opens the app and what they need to do. Third, a clear success check: how you will know it works, such as "a new lead appears in the list within a few seconds." If you can't name all three, write them down before prompting the AI.
What vibe coding gets right, and where it breaks
The strengths

- Speed: you can go from idea to a working preview in an afternoon rather than weeks.
- Cheap testing: you can check whether an idea is worth pursuing before you hire anyone.
- No language to learn first: you can start with the problem and learn the technical side as you need it.
The weak spots
- Unreviewed code: the app may work on the surface while containing logic errors you never see.
- Missing security checks: AI-generated apps can skip permission checks, input validation or safe handling of passwords and payments.
- Bugs that appear only with real data: a form that works with three test rows can fail with 3,000 real ones or with unusual characters in names.
For a structured way to examine AI-built code before you rely on it, use the review checklist for vibe coding apps.
Is vibe coding safe?
Vibe coding is safe enough for prototypes, internal tools and low-stakes personal projects, as long as you treat the output as a draft. It becomes riskier when the app handles customer personal data, payments, medical or legal information, or login systems for many users. In those cases, have someone with security experience review the app before launch. Recent public incidents involving AI-built software show why this matters; for one example and what it teaches, read the lessons from the Grok Build leak.
Warning signs to check before launch
- The app shows data to people who shouldn't see it. Log in as a second user and check.
- Forms accept anything, including empty fields, very long text or special characters.
- Passwords or API keys appear in the page source or in plain text in the database.
- The app works only in the one browser you tested.
- You can't explain, in your own words, what happens when someone submits the main form.
If you see any of these, stop and fix them before sharing the link.
How to start vibe coding without coding
You can try this this week with one small project.
- Pick one small, real problem. Choose something you do repeatedly, such as tracking client requests or taking bookings. Avoid "build the next big app."
- Write a one-paragraph description. Include the users, the data and the main action. For example: "Freelance designers track client requests in a spreadsheet. I want a form clients fill in, a list showing each request's status, and a way to mark requests as done."
- Build a first version and test it with three real tasks. Submit a request, change its status, and check that a second browser sees the same data.
- Keep a record of each version. Save each working state so you can roll back if a later change breaks something.
If you are not sure where to begin, an AI app builder is a practical place to start, because it combines the description, preview and version steps in one place. You can begin on the free plan and check pricing when you need more projects or credits.
Build it with FromScratch
FromScratch is an AI app builder. You describe the app in a chat, and an agent builds a React front end, a FastAPI back end and a database. The app appears in a live preview within minutes, and you refine it by continuing the conversation. Here is how to build a small vibe coding quiz with it:
- Sign in on the FromScratch site and start a new project.
- In the chat, paste the prompt below exactly as written.
- Open the live preview and click through all five questions. Check that every answer shows an explanation and that the restart button resets the score.
- Refine in small steps if needed, for example: "make the results screen show the score as a percentage."
- Every change is saved, so you can restore an earlier version if a later edit breaks something.
- When you are happy with it, publish to a public URL, or connect your own custom domain.
Build a simple one-page quiz app that asks five questions about vibe coding and shows a score at the end. Each question should have four answer choices, and the app should show a short explanation after each answer. Include a button to restart the quiz and a results screen that suggests a next step for beginners.
FAQ
Do I need to know how to code to vibe code?
No, but you need to judge whether the result does what you asked. Test each feature with realistic data, and get a technical review before you handle customer data or payments.
Is vibe coding the same as no-code?
No. No-code tools rely on fixed building blocks you arrange visually. Vibe coding generates real software from your description, so you can change the code, the data model and the logic later.
Can I vibe code a business app?
Yes, for many internal tools and prototypes. Review the app carefully before you rely on it for customer information, payments or logins.
Who coined the term?
Andrej Karpathy popularized "vibe coding" in early 2025. The term now covers any workflow where you build software mainly by describing it to an AI.