What is vibe coding?
Vibe coding is building software by describing what is wanted to an AI model in plain language and accepting the generated code without reading it, judging the result only by whether the program appears to work.
AI researcher Andrej Karpathy coined the term in a post on X on 2 February 2025. He described talking to a coding tool by voice, accepting every change without reading the diffs, pasting error messages back in without comment, and letting the code grow beyond his own understanding. He called it fine for throwaway weekend projects.
The phrase spread fast. Collins named "vibe coding" its Word of the Year on 6 November 2025, defining it as using AI prompted by natural language to help write computer code. In that wider sense it often gets used for any AI-assisted programming, which blurs a useful distinction.
Vibe coding versus reviewed AI coding
Simon Willison's post "Not all AI-assisted programming is vibe coding" (March 2025) draws the line most engineers now use. Vibe coding means not reviewing the code. Using a model to write code that is then read, tested and understood is ordinary software development with a faster typist. His own rule is not to commit code he could not explain to someone else.
The distinction matters because the two produce very different risk. The same model, used the same way, gives a disposable prototype in one case and maintainable software in the other; the difference is whether a person or a test suite checks what it wrote.
Where vibe coding works
Prototypes
Testing whether an idea or interface is worth building, where the code will be thrown away.
Personal tools
Small scripts and apps used by the person who made them, with little data at stake.
Exploration
Getting a feel for an unfamiliar library or API before writing the real version.
Limits for production software
Unreviewed security
Code nobody has read can expose secrets, skip authorisation or trust user input. Working is not the same as safe.
No one understands it
Karpathy's own description notes the code grows beyond comprehension. The next change, or the next incident, starts from zero.
Workarounds instead of fixes
When a bug resists, vibe coding routes around it. Those detours pile up as hidden defects.
No tests, no contract
Without tests or a written spec, there is no way to tell whether a later change broke something.
From prototype to production
Tool makers now pitch the path out of vibe coding explicitly. AWS introduced Kiro (July 2025) as a way to take prompt-built prototypes into production through specifications, and GitHub's Spec Kit announcement (September 2025) called vibe coding fine for quick prototypes but less reliable for mission-critical work or existing codebases. Motiontic's engineers treat a vibe-coded prototype as a useful sketch: they keep what it proved about the product and rebuild the code with tests, review and a written spec.
Vibe coding compared with agentic coding under review
| Vibe coding | Agentic coding under review | |
|---|---|---|
| Who reads the code | Nobody | Tests first, then an engineer |
| How success is judged | The program appears to work | Tests pass and a reviewer approves |
| Speed to first version | Very fast | Fast |
| Maintainability | Low, the code is not understood | Comparable to hand-written code |
| Fits | Prototypes and throwaway tools | Software that people and data depend on |
Common questions
Who coined the term vibe coding?
Andrej Karpathy, in a post on X on 2 February 2025. Collins later named it Word of the Year for 2025.
Is vibe coding the same as using AI to code?
Not in its original sense. Vibe coding means accepting AI-written code without reading it. Writing code with AI and then reviewing and testing it is ordinary software development.
Can a vibe-coded app go to production?
Rarely as it stands. It usually needs a review for security, tests, and often a rewrite of the parts nobody understands before it can safely hold real users or data.
Related
Agentic coding
Engineers delegate whole tasks to a coding agent and spend their time on direction and review.
Spec-driven development
Write down what the software must do first, then let the coding agent plan and build against it.
Coding agent
An agent with a shell and a repository: it reads, edits, runs the tests and tries again.
Claude Code vs Cursor
Anthropic's coding agent and the SpaceX-owned AI editor compared on plans, models and agent tooling.
AI product development
New products and major AI features, built from scratch alongside the client's team.
Sources
- Andrej Karpathy, post on X introducing "vibe coding" (2025-02-02)
- Collins Language Blog, Collins Word of the Year 2025 (2025-11-06)
- Simon Willison, Not all AI-assisted programming is vibe coding (2025-03-19)
- Kiro blog, Introducing Kiro (2025-07-14)
- GitHub Blog, Spec-driven development with AI: Get started with a new open source toolkit (2025-09-02)
Running agents that have to hold up?
Describe the agent, the systems it touches and what it must never get wrong. The first call is with a lab engineer, who will say plainly whether it is ready for production, and whether Motiontic is the right team.