What Is Vibe Coding? Definition, Origins, and Where It Breaks Down

Vibe coding means describing what you want to an AI tool, running the result, and asking for changes without closely reading or understanding the generated code.

Today, the term is also used more loosely for almost any AI-assisted programming, which causes much of the confusion around it.

That difference matters. The original workflow may suit low-stakes experiments, but it becomes risky when an app handles personal data, payments, or work that other people depend on.

This guide explains where vibe coding came from, how it works, what it is useful for, and where it starts to break down.

Vibe Coding Meaning in Simple Words

In simple words, vibe coding means telling an AI what an app should do and judging the result by what appears on screen.

Instead of writing or reviewing each line of code, you keep prompting until the app seems to work.

For example, someone might ask an AI to create an expense tracker, run the generated app, and report any problems through more prompts. If that person accepts each change without examining the code, that is vibe coding.

Someone else could use the same AI tool but review every change, test the app, and understand how it works. That is AI-assisted software development rather than vibe coding in its original sense.

Why Is It Called Vibe Coding?

Andrej Karpathy introduced the phrase in a post on X on February 2, 2025.

He described letting the AI take over, accepting changes without reading the code, and responding to errors with more prompts.

The name reflects that loose, results-first approach: follow the “vibe” of whether the app appears to work instead of examining how the code works.

Karpathy described it as suitable for throwaway weekend projects, not as a standard process for production software. Merriam-Webster’s word history records the term’s origin and date.

How Fast the Term Spread

The phrase moved into mainstream use within weeks. Merriam-Webster added slang coverage in 2025, and Collins named “vibe coding” its 2025 Word of the Year.

As the term spread, its meaning also widened.

Some people now use “vibe coding” for almost any software built with AI, even when the generated code is carefully reviewed.

The Two Meanings of Vibe Coding

Vibe coding now has two common meanings.

The original meaning is narrow: someone accepts AI-generated code without reading it and judges the app mainly by what happens on screen.

The broader meaning covers almost any software built with AI, including work that engineers review and test carefully.

Check which meaning a source is using before trusting its claims. Evidence about reviewed, professional development does not prove that unreviewed code is safe.

Simon Willison’s Test for Vibe Coding

Simon Willison offers a practical test: if someone reviews the generated code, tests it thoroughly, and can explain how it works, the result is software development.

If the code is accepted without being understood, the process fits the original definition of vibe coding. The dividing line is responsibility, not whether AI helped write the code.

How Karpathy Updated the Definition in 2026

By 2026, Karpathy was drawing a clearer line between casual prompting and professional use of AI agents.

In his April 2026 Sequoia Ascent summary, he said vibe coding “raises the floor” by allowing more people to create software. Agentic engineering “raises the ceiling” by using AI agents while preserving correctness, security, taste, and maintainability.

The AI may produce most of the code, but a person still defines the goal, checks the work, and takes responsibility for the result.

Why the Term Is Used as Criticism

“Vibe coding” is sometimes used as a put-down for careless AI-generated work. The criticism usually targets practices such as:

• Accepting generated code without reviewing it
• Treating visible results as proof that the app works
• Shipping code without checking security or maintainability

However, the term is also applied unfairly to careful AI-assisted development.

Andrew Ng argued in 2025 that the name can make the work sound effortless, even though directing, checking, and debugging AI-generated code can require intense concentration.

How Vibe Coding Works

Vibe coding follows a simple feedback loop: describe what you want, let the AI generate code, run the result, report what went wrong, and repeat.

The person judges how the app behaves rather than reviewing how it was built.

1. Describe the Outcome

Explain the goal, users, inputs, expected result, and important rules in plain language. The prompt describes what the app should do rather than how to code it.

For example: “Create an expense-request form. Require a category, amount, receipt, and manager approval for requests above $500.”

2. Let the AI Generate a First Version

An AI code editor, browser-based app builder, coding agent, or chat assistant creates the files or changes an existing project.

Some tools can work across the entire project, edit several files, run commands, and respond to errors.

3. Run the Result

Open the app and click through it. Try normal inputs, missing fields, unusual values, and repeated actions to see what breaks or behaves incorrectly.

A page that loads successfully does not prove that the code is correct, secure, or easy to maintain.

4. Feed Back What Broke

Paste the error message or describe the incorrect behavior, ask the AI to fix it, and run the app again.

This is where vibe coding can become difficult. A visible problem may disappear while the fix creates another issue elsewhere. Without reviewing the code or running broader tests, the person may never see that tradeoff.

The Tools Used for Vibe Coding

Vibe coding can happen with several types of AI tools:

• AI code editors: Cursor can suggest code and make changes within an editor.
• Browser-based AI app builders: Lovable can generate web applications from prompts.
• Agentic coding tools: Claude Code can read project files, edit code, and run commands.
• General chat assistants: ChatGPT can draft code, explain errors, and suggest changes.

These categories often overlap. The tool itself does not determine whether the work is vibe coding. What matters is whether the generated code is reviewed, tested, and understood.

A Simple Vibe Coding Example

Suppose an operations manager wants a small internal expense tracker.

1. First prompt: “Build a form that records an employee name, amount, category, receipt, and approval status.”
2. AI-generated result: The tool creates a form and a table of submitted requests.
3. Failed test: The manager enters a negative amount, and the form accepts it.
4. Follow-up prompt: “Reject amounts at or below zero and show a clear error beside the amount field.”

The visible problem is fixed, so the manager continues testing. This is vibe coding because the result is judged by how it behaves rather than by reviewing the generated code.

However, the test does not show who can view receipts, change approvals, or access the stored data. Those questions require code inspection, access-control testing, and broader engineering review.

What Vibe Coding Is Used For

Vibe coding works best when mistakes have limited consequences and the code will not require long-term maintenance. It can provide fast feedback, but it does not prove that the result is ready for production.

Good Fits for Vibe Coding

• Prototypes: Test whether an idea is useful before investing in a full build.
• User interface (UI) mockups: Explore screens and interactions before engineering begins.
• Personal tools: Automate a low-risk task for one person.
• One-off scripts: Transform non-sensitive data for a temporary need.
• Small internal experiments: Trial a workflow with a few informed users.
• Learning projects: Explore how interfaces, data, and application logic connect.

When to Stop Vibe Coding

Stop and introduce formal review before an app stores other people’s personal data, processes payments, controls safety-critical actions, becomes important to daily work, or requires long-term maintenance.

At that point, define the requirements, inspect the code, test failure cases, review dependencies, set access controls, and assign responsibility for maintaining the app.

Is Vibe Coding Bad?

Vibe coding is useful for low-stakes experiments and risky for software that handles sensitive data or affects other people.

What People Get From It

The main benefits are faster prototypes, a lower barrier for non-coders, and quick feedback on an idea.

Speed is not guaranteed. A 2025 Model Evaluation & Threat Research (METR) study found that 16 experienced developers completed 246 tasks 19% more slowly with early-2025 AI tools, despite believing they were 20% faster.

The result covers experienced developers working in familiar, mature projects, not every user or newer tool.

Common Failure Modes

Unreviewed code can contain hidden bugs, weak access controls, insecure defaults, tangled structure, and dependencies the builder never checked.

A 2026 security study audited 200 deployed apps from a collection of 9,041 projects. It found that 91% had at least one vulnerability, while 65.77% of the validated findings were rated Critical or High.

This was a selected sample in a recent preprint, not proof that 91% of all AI-assisted software is vulnerable.

Terms People Confuse With Vibe Coding

The main difference is who reviews the code and takes responsibility for it.

Is Vibe Coding a Real Job?

Vibe coding is a technique, not a standard occupation.

It may be used within software, product, design, or operations roles, but employers pay for the person’s responsibilities and results.

In the 2025 Stack Overflow Developer Survey, 72.2% of respondents said vibe coding was not part of their professional work, while another 5.3% gave an emphatic no.

What Y Combinator’s 2025 Numbers Show

Y Combinator (YC) reported that one-quarter of its Winter 2025 batch had codebases that were 95% generated by AI.

However, AI-generated code does not show how much the founders understood, reviewed, or tested. YC’s Jared Friedman said these founders were highly technical and could have built the products themselves.

How Much Do Vibe Coders Make?

There is no reliable salary category for “vibe coders.” Pay depends on the actual role, experience, and location.

For comparison, the U.S. Bureau of Labor Statistics reported a median annual wage of $135,980 for software developers in May 2025. That is not a vibe-coder salary estimate.

Before You Try Vibe Coding

Use this checklist for a first project:

• Choose something low-stakes and easy to reverse.
• Define what a correct result should do.
• Keep real passwords, secrets, and customer data out of prompts and test environments.
• Save versions so a working state can be restored.
• Test normal inputs, invalid inputs, permissions, and failure cases.
• Ask an experienced reviewer to inspect the code before anyone else depends on it.

How Modelence Fits After Vibe Coding

Vibe coding can produce a first version, but shipping it requires more work.

Before production, the code should be reviewed, tested, secured, and prepared for a suitable hosting environment. Modelence Cloud can fit at this later stage as an app hosting platform.

Modelence does not replace code review, security testing, or maintenance planning.

The team remains responsible for checking the application’s code, data access, dependencies, and security before launch.

Once the application has passed these checks, confirm hosting compatibility and consider Modelence Cloud.