What Is Vibe Coding? Building Apps by Talking to AI

Where the term vibe coding came from, how AI turns plain-language prompts into apps, what it suits, and what to check before others use it.

You have probably seen a post like this in your feed: "Built a booking app for a barber shop in one night without writing a single line of code myself." Someone replies, "That's vibe coding." The term is now used by tech people and non-tech people alike, but what is it exactly, and how is it different from ordinary AI-assisted coding? The answer gets much clearer once you understand the mechanism behind it.

What is Vibe Coding?

The term comes from Andrej Karpathy (former director of AI at Tesla and a founding member of OpenAI), who posted on 6 February 2025 about a way of working where you "fully give in to the vibes" and "forget that the code even exists". You talk to the AI, say what you want, look at the result and ask for changes, without reading the code line by line. Collins Dictionary later named "vibe coding" its Word of the Year 2025, defining it as the use of artificial intelligence prompted by natural language to assist with the writing of computer code.

Simon Willison, a developer who has written about this in detail, draws a useful line: vibe coding in its original sense means building software with an LLM without reviewing the code it writes. If you review, test and can explain what the code does, that is software development with AI assistance, not vibe coding.

Behind the scenes: how AI "writes the app" for you

1. Your prompt becomes a spec

What you type, for example "make a page where customers pick a date and time to book", is sent to an LLM together with the project's context (existing files, page structure, earlier messages). The model breaks it into sub-tasks: a form page, a table to store bookings, a confirm button.

2. An agent edits the files

Today's tools do not just reply with text. They work as coding agents: creating and editing real project files, installing libraries, running commands and checking the result. For a deeper look at agents, read What is a Coding Agent.

3. Preview, then loop

You see the app immediately, click around, fill in forms, and ask for changes when something is off. The loop is very short: "think, write, test" that used to take days now takes minutes. That is why it feels like "chat and get an app".

What vibe coding is good for

• Prototypes and MVPs: test whether an idea works before investing in a team
• Personal or small-team tools: data collection forms, pricing calculators, simple dashboards
• Landing pages and market tests: test a sales message or collect sign-ups
• Learning: let AI build something, then ask how each part works. A more fun way to learn programming

Be much more careful with systems that store personal data, take payments, or that many people depend on. Those always need review from someone who understands the code.

Example: a café in Chiang Mai wants a loyalty system

Imagine the owner of a small café who wants to drop paper stamp cards. She opens a vibe coding tool and types: "Make a web page where staff enter a customer's phone number, add points, and show when they reach 10 cups for 1 free cup." After a few rounds she has a working page. She uses it at the counter for a week, notices staff often mistype numbers, and asks the AI to add phone number format validation. That is the power of vibe coding: turning a real problem into a tool quickly.

What she still has to think about: customer phone numbers are personal data. Who can see them? If the link leaks, can outsiders see the customer list? The AI will not always ask these questions for you.

Vibe coding lets you build faster, but responsibility for what you build is still yours.

Things to watch out for

• Secrets: never paste API keys or passwords into the chat or into front-end code. Store them where the tool provides a dedicated place for secrets
• Real data: use fake data while experimenting. Before storing real customer data, set access permissions first and keep Thailand's PDPA in mind
• Understanding: Simon Willison's personal rule is not to commit code he could not explain to someone else. If other people will use the app, keep asking the AI until you can explain it, or have someone experienced review it
• Costs: if the app calls a pay-per-use API, always set a spending cap

How to get started

1. Pick a small problem you hit every week, such as a form you fill in repeatedly or a calculation you redo in Excel
2. Write what you want in plain language: who uses it, what for, and what is on the screen
3. Build it with a browser-based tool and test with fake data
4. Every time the AI changes something, ask "what did you change and why" to build up your understanding
5. Before sharing with others, always check permissions and data first

For a hands-on, step-by-step example, read Build an app with AI in one day, or learn in a structured way with the course Vibe Coding: build an AI web app with Lovable on Plants.work, which brings together courses from learning partners so you can choose by career goal.