Vibe coding means building software by describing what you want to an AI model in plain language and accepting the code it writes, often without reading that code closely. For founders, it can turn a weekend idea into something people can click on by Sunday evening. It has also produced plenty of apps that fall over the moment real customers arrive. This guide covers where the term came from, what it looks like in 2026, and how to decide whether it belongs in your business.
What is vibe coding, in one paragraph
You open an AI tool and type something like “a booking page for my yoga studio with a calendar and email confirmations”. The tool writes the code, runs it and shows you a preview. When something looks wrong, you describe the problem and it tries again. You judge the result by whether it works on screen, not by reading the code. That last part separates vibe coding from the wider practice of AI-assisted programming, where a developer uses the same tools but reviews every change before it ships.
The difference matters because the two carry different risks. A developer using an AI assistant is still accountable for the code. A founder who is vibe coding is trusting the model with decisions about security, data and structure that they may not know were being made.
Where the term came from
The phrase comes from Andrej Karpathy, an AI researcher who was a founding engineer at OpenAI and later director of AI at Tesla. In a post on X on 2 February 2025, he described a way of building where you “give in to the vibes” and stop paying attention to the code itself. Merriam-Webster, which now has an entry for the term, dates its first known use to that post.
It spread quickly. In November 2025, Collins Dictionary named vibe coding its word of the year, describing it as software development that turns natural language into computer code using AI (Gulf News, November 2025). Within a year, a phrase from one social media post had become a dictionary entry and the name of a product category.
What it looks like in 2026
The tools have moved on a lot since early 2025. Most fall into two broad groups.
- App builders take a plain-language prompt and produce a working web app with a preview you can share. Examples include Lovable, Replit Agent and Vercel’s v0. Lovable’s documentation says projects produce editable code that can be synced to GitHub, GitLab or Bitbucket, and Replit’s says its Agent saves checkpoints so you can roll back.
- Coding agents work inside an existing codebase. Claude Code, Cursor, OpenAI Codex and GitHub Copilot’s cloud agent read files, run commands and propose changes. They suit people who can review what the agent produces.
Founders usually start in the first group and engineers mostly live in the second. Plenty of teams use both: a founder sketches a working prototype in an app builder, then a developer moves the code into a proper repository and carries on with an agent and a test suite.
Where it works well
Vibe coding is very good at getting from an idea to something people can click. That changes early conversations. Say a founder in Nairobi wants to test a delivery-tracking idea with ten shop owners. A clickable prototype built in an afternoon gets better feedback than a slide deck, and it costs very little to throw away if the idea turns out to be wrong.
It also suits internal tools with a small, known group of users: a stock tracker for a trading company in Dubai, a lead list for a sales team in Pune, a form that replaces a shared spreadsheet. If the tool breaks, the people affected sit in the same office and will tell you the same day.
Where it goes wrong
The trouble starts when a prototype quietly becomes the product. AI models write code that runs, and running code can still be unsafe. In a July 2025 study covering more than 100 large language models, Veracode found that 45% of the coding tasks it tested produced code containing a security flaw from the OWASP Top 10 (Veracode, 2025).
Models also make things up. A paper presented at USENIX Security 2025 found that code-generating models recommended software packages that do not exist, at an average rate of at least 5.2% for commercial models and 21.7% for open-source ones (Spracklen and others, 2025). Attackers can register those invented names and fill them with malicious code.
None of this shows up in a preview window. The app looks finished. The problems sit where a non-developer would not think to look: who can read the database, where the API keys are stored, and what happens when two people press the same button at once.
How founders should think about it
Treat vibe-coded software the way you would treat an architect’s sketch. It is excellent for deciding what to build and for showing investors or early customers what you mean. You would not hand it to paying customers, or trust it with their personal data, until someone has checked it.
A practical rule: vibe code freely for prototypes and internal tools, then bring in review before anything touches money, customer data or a public launch. Our insights section has a separate checklist covering what to check before shipping AI-written code. If you already have a working prototype and want an outside view of what it would take to make it production-ready, that is the kind of question our technology advice service is set up to answer.
Frequently asked questions
Who coined the term vibe coding?
Andrej Karpathy, an AI researcher who was a founding engineer at OpenAI and later director of AI at Tesla, used it in a post on X on 2 February 2025. He described building software by talking to an AI and not paying close attention to the code. Collins Dictionary named it word of the year in November 2025.
Is vibe coding the same as using Cursor or GitHub Copilot?
Not quite. Those tools can be used for vibe coding, but in professional teams they are normally paired with review, where someone reads and tests each change before it ships. Vibe coding usually means accepting what the AI produces because it seems to work on screen. The tools can be identical; the difference is how much checking happens before release.
Can I launch a real business on a vibe-coded app?
Some founders do, but the risk grows with every paying customer and every piece of personal data you hold. Use vibe coding to test the idea and win early users. Before you take payments or store customer data at any scale, have someone review security, data handling and who owns the code and accounts.