What is vibe coding?
Vibe coding is building software by describing what you want in plain English and letting an AI write the code. You do not write the code yourself. You describe the thing, the AI builds it, you check whether it works, and you keep going until it does.
A longer answer, still in plain English.
For most of computing history, building software meant learning a programming language and typing it out by hand, one line at a time. If you had not spent years doing that, you could not build. You had to hire someone, or give up on the idea.
Vibe coding removes that step. You open an AI tool, you type what you want in ordinary words, and it writes the code for you. You read the result, tell it what is wrong, and it fixes it. The loop repeats until the thing you imagined is sitting there working.
The word “vibe” is doing real work in that phrase. You are steering by feel. You are not tracking every line of code the AI wrote, the same way you are not tracking every gear turning inside a car when you drive one. You care whether it goes where you pointed it.
The two things you still have to do
Vibe coding removes the typing. It does not remove the thinking. Two jobs stay yours, and they are the two that decide whether you end up with something real.
An AI will build exactly what you describe, including the vague parts. A fuzzy idea produces a fuzzy app. The people who get good results are not the best prompters, they are the ones who got clear on the idea before they started typing.
AI is confidently wrong sometimes. It will hand you a button that does nothing and tell you it works. You do not need to read the code to catch this. You need to click the button.
One post in February 2025.
The phrase was coined by Andrej Karpathy, a founding member of OpenAI and the former director of AI at Tesla. On 2 February 2025 he described a new way of working where you “fully give in to the vibes” and stop thinking about the code as code.
He was describing something he had noticed in his own work. The AI models had quietly crossed a line. They had become good enough that for small projects you could stop reading what they wrote and just keep talking to them until the thing worked.
The phrase spread far past programmers. In November 2025, Collins Dictionary named vibe coding its Word of the Year, which is a strange fate for a term that was nine months old and started as a post on X.
It named something people were already quietly doing.
By early 2025 a lot of people who could not code were already opening AI tools and asking them to build things. They just did not have a word for what they were doing, or permission to think of it as real. The term gave both at once.
A real example, start to finish.
Say you run a small bakery and you are tired of taking custom cake orders over Instagram DMs. You want a simple form on a page. Here is the whole process.
You open an AI tool that can write and run code, and you type this:
I run a small bakery. I want a one page website with a form where customers order a custom cake. The form should ask for their name, email, the date they need it, the size, and a box for special requests. When they submit it, email the order to me. Make it look warm and simple, not corporate.
Ninety seconds later you have a working page. You look at it. The date picker lets people choose yesterday, which is not useful. So you say so:
The date picker lets people pick dates in the past. I need at least three days notice, so block anything sooner than that.
It fixes it. You keep going like this. Maybe six or eight rounds. Then you put it on the internet at a real web address and start sending people there instead of your DMs.
At no point did you read a line of code. That is vibe coding. The whole thing is a conversation where one side happens to be able to build.
Four things vibe coding is not.
The term gets used loosely, and a few misunderstandings do real damage to people just starting out. These are the ones worth clearing up.
No-code tools like Bubble or Webflow give you a set of blocks to arrange, and you can only build what the blocks allow. Vibe coding produces genuine source code, so nothing is off the table. The tradeoff is that you own real code, which is more powerful and slightly scarier.
You will hit walls. Things will break in ways that make no sense. Your first build will take longer than you expect. The difference is that being stuck now costs you an afternoon rather than a six-month course.
It is very good at small, self-contained things. Banking systems, medical records, and anything holding a lot of strangers' personal data still need people who genuinely understand the code. Knowing which side of that line your idea sits on is part of the skill.
Most things built this way are small and personal. A tool that sorts your invoices. A tracker for your running club. A page that does one annoying task you do every Monday. None of that is a business, and all of it is worth building.
What you can build, and where it gets hard.
Anyone telling you there are no limits is selling something. Here is the honest map, based on what people actually ship.
Works well
Landing pages and personal sites. Forms that collect something and send it somewhere. Internal tools that save your team a repetitive job. Dashboards that pull numbers into one place. Small apps with a clear job. Prototypes you need in front of someone this week. Automations that move data between two services you already pay for.
Gets harder
Anything with lots of users at once. Payments, past the simple checkout that a service like Stripe hands you. Apps where several people edit the same thing at the same time. Anything that has to be fast at large scale. You can still start these with AI. You will want a developer before real money runs through them.
Get help
Health records, financial accounts, anything holding sensitive information about people who are not you. Not because AI cannot write the code, but because AI writes security holes with exactly the same confidence it writes working features, and you will not spot the difference by clicking around.
Build it yourself until someone else could get hurt.
If the worst case of a bug is that you are annoyed, build it yourself. If the worst case is that someone loses money or has their private information exposed, bring in a person who reads code. That single line covers almost every decision you will face.
Mostly people who were told they could not.
The people getting the most out of this are usually not programmers. They are people with a specific problem and no way to solve it until now.
Small business owners who need one tool that nobody sells. Founders who need something real to show an investor before the money runs out. Teachers, nurses, and accountants who know exactly what would fix their week. People who have been carrying an idea around for years, waiting for a developer who never became affordable.
Career programmers use it too, and often. For them it is a speed increase. For everyone else it is the difference between having an idea and having a thing.
Words you will hear, translated.
Most of the intimidation in this space is vocabulary. Here is the jargon you will run into, in ordinary words.
- Prompt
- What you type to the AI. That is it. A prompt is just a message.
- LLM
- Large language model. The kind of AI behind Claude and ChatGPT. It is the thing writing your code.
- Deploy
- Putting your thing on the internet so it has a real web address other people can visit. Before you deploy, it only exists on your screen.
- Frontend
- The part people see and click. Buttons, text, colours, layout.
- Backend
- The part nobody sees. Where information is stored and the actual work happens.
- Database
- Where your app remembers things after someone closes the tab. A spreadsheet that your app can read and write.
- API
- How two pieces of software talk to each other. When your app sends an email through another service, it uses that service's API.
- Repo
- Short for repository. The folder holding all your project's code, usually with a history of every change.
- MVP
- Minimum viable product. The smallest version worth showing anyone. Build this, not the full thing in your head.
- Shipping
- Releasing it. Putting it in front of real people instead of tinkering with it forever.