Vibe coding is a way of making software where AI writes the code from a plain-language description and the person barely reads it, judging by the result. Andrej Karpathy coined the term on February 2, 2025, and Collins named it its Word of the Year 2025.
Where the word came from
The term comes from AI researcher Andrej Karpathy. On February 2, 2025, he posted on X (the post) about "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." In the same post he says he hits Accept All, doesn't read the diffs, pastes error messages back with no comment, and that it's "not too bad for throwaway weekend projects."
The meaning has widened since. Today people often call any work where AI writes the code from a plain description vibe coding, while the original point was that nobody reads that code.
What the dictionaries say
On November 6, 2025, Collins named vibe coding its Word of the Year 2025 (Collins blog), describing it as the use of artificial intelligence prompted by natural language to write computer code. Merriam-Webster lists vibe coding as a noun, the act or practice of using an AI system to generate code in a given programming language, and dates its first known use to 2025 (Merriam-Webster).
| Date | What happened |
|---|---|
| February 2, 2025 | Andrej Karpathy's post on X, first use |
| November 6, 2025 | Collins names vibe coding Word of the Year 2025 |
| February 3, 2026 | Xcode 26.3 adds coding agents, Anthropic's Claude Agent and OpenAI's Codex |
| April 28, 2026 | The App Store only accepts builds from Xcode 26 or later |
| June 8, 2026 | Latest App Review Guidelines, rule 2.1(a) asks you to test on a device before submitting |
Per Karpathy's post, Collins, Apple Newsroom and the App Review Guidelines, checked October 16, 2026.
Vibe coding vs agentic coding vs regular coding
In regular programming a person writes and reads every line. In vibe coding in the original sense you describe what you want, accept whatever the AI produced and judge by what you see on screen. Working with an AI agent sits in between. The agent writes the code, builds the project and fixes errors, while you set the task, review the changes and decide what stays.
That is how Apple describes the agents in Xcode 26.3, released in February 2026 (Apple Newsroom). They search documentation, explore the file structure, update project settings and check their own work with Xcode Previews and repeated builds.
Where it lives in Xcode
You add an agent in Xcode › Settings › Intelligence with the Get button next to it under Agents (Xcode Help). The coding assistant opens with the Coding Assistant button or Command-0. The agent's changes show up next to the conversation, and History rolls them back, which requires a Git repository (Xcode Help).
What it looks like for an iPhone app
- Describe the app in plain words, what's on the first screen, what people do and what they pay for. How to write that up is covered in how to brief an AI agent.
- The agent creates an Xcode project in SwiftUI and writes the screens.
- Xcode builds the project, and the agent reads the build errors and fixes the code until it compiles.
- You test the app on the simulator and on your own iPhone, as the App Store rules ask.
- A build goes to a few people through TestFlight, and they find what you missed.
- Then the product page, purchases and submission in App Store Connect.
Common mistakes
- Accepting everything unread where money and data are involved. Karpathy was talking about weekend projects. In an app with purchases, unread code means untested payments and untested handling of personal data.
- Judging by one run. A clean build doesn't mean every screen, the purchase and restore purchases all work.
- Working without Git. An agent edits many files at once, and without history you can't roll back a bad change.
- A paid AI service key inside the app. A key baked into the build can be pulled out of it, and you pay for other people's requests.
Appto works with Claude Code and Codex on your own subscription and takes the agent through the whole path, from the niche and design to the code, screen checks on the simulator, the App Store page and App Store Connect. More on the code page.
Questions
Who coined the term vibe coding?
Andrej Karpathy, an AI researcher, in a post on X on February 2, 2025. Collins and Merriam-Webster both trace the word to that post.
Is vibe coding good or bad?
It depends on the job. It works for a prototype or a weekend project, which is what Karpathy had in mind. In an app that takes payments or holds personal data, someone still has to check the code.
What is the difference between vibe coding and agentic coding?
In the original vibe coding you don't read the code and accept every change. With an agent you set the task and review the result while the agent builds the project and fixes errors, the way Apple described agents in Xcode 26.3 in February 2026.
Can you vibe code an app for the App Store?
Yes, as long as it's built with Xcode 26 or later, tested on a device and follows the App Store rules. The rules are the same for any code, whoever wrote it.
Is it vibe coding, vibecoding or vibe-coding?
All three appear. Merriam-Webster lists vibe coding, and less commonly vibecoding and vibe-coding.
Build your own app in Appto
Appto checks the niche on fresh App Store data, designs the screens, writes the code, builds the App Store page and submits to Apple from your account or shows you what to press in App Store Connect. Runs on your own Claude Code or Codex.
Download for Mac