Skip to content

Vibe coding

| Article: 4 min read | AI in e-commerce

Vibe coding is building programs where a person describes the task in ordinary language, AI writes the code, and the person accepts the result almost without checking it and goes by whether it "looks like it works", adding new wishes in the chat. For quickly testing an idea this is handy. For a store that takes money and holds buyers' data, it is a risk that needs to be understood.

Where the term comes from

The term was proposed by AI researcher Andrej Karpathy in an X post on 2 February 2025: a style in which you "fully give in to the vibes" and forget the code exists. Collins English Dictionary named "vibe coding" its word of 2025 (Wikipedia, Vibe coding). This is the history of the term, not data.

An important distinction. If a developer reads, tests and understands the code AI wrote, that is AI-assisted development. Vibe coding means the code is accepted without being understood. Simon Willison, quoted by Ars Technica, said that "vibe coding all the way to production code is clearly risky" (ibid.).

When it's fine

  • Prototype and idea testing. Show the team what a delivery calculator, a size-picker form or a landing page could look like before asking developers to build it.
  • One-off scripts for yourself that don't touch customer data: rename files, merge two spreadsheets.
  • Learning. See what is possible and understand what questions to ask specialists.
  • An internal tool without access to payments and personal data, if a few people use it and a loss isn't critical.

When it's dangerous

  • A store or account area that takes payment or stores buyers' data. Here a mistake in access or data format turns into a leak or lost money.
  • Anything connected to real keys, databases and payments. Wikipedia names among the cases: on the Lovable platform in May 2025, 170 of 1,645 checked apps had a problem exposing personal data; in July 2025 a Replit AI agent deleted a company's production database despite an explicit ban on making changes (Wikipedia, Vibe coding). These are examples, not market statistics.
  • A project that will live for years. Code nobody understands is more expensive to change. There is a separate article on this: technical debt.

What research says

  • Security. In July 2025 Veracode tested more than 100 models: in 45% of test samples the code contained vulnerabilities from the OWASP Top 10, and newer models wrote working code better but not more secure code (Veracode, 2025). In October 2026 Veracode reported the first noticeable improvement: the overall security pass rate rose from 56% to 62%, but the weakest models remain at 44–50%, and Veracode advises combining AI tools with independent security review (Veracode, 2026). All of these are results of test tasks, not the share of real code in production.
  • Speed is not guaranteed. In a randomized METR study (16 experienced developers, 246 tasks in familiar projects, early 2025), tasks with AI took 19% longer, although the developers themselves believed they were faster. The authors caution that the result applies to specific conditions, and the page itself is marked as outdated and points to a newer 2026 study (METR, 2025). This is about experienced engineers, not beginners.
  • Developer practice. In the Stack Overflow Developer Survey 2025, 72% of respondents said they don't do vibe coding, another 5% "definitely not"; about 15% answered "yes" to varying degrees (Stack Overflow, 2025).

How to work more safely

  1. Decide what you are testing for speed and what will go on sale. A prototype must not turn into a store "by itself".
  2. Don't paste real keys, passwords and customer databases into the chat.
  3. Ask a specialist to review the code and configuration before launching to people.
  4. Keep backups and a separate environment for experiments.
  5. Record what exactly was created by AI so the next developer knows what they are working with.

Common mistakes

  • Taking a working prototype for a finished product.
  • Thinking that "works" means "secure". Code can work and be vulnerable at the same time.
  • Giving an agent access to the production database.
  • Having no backup and no change history.

FAQ

What is vibe coding in plain words? It is programming by conversation with AI: you describe what you need and AI writes the code. You accept the result without reading it.

Can an online store be built with vibe coding? A prototype, yes. For sales with real payments and buyers' data you need a specialist's review: security, payment and backups. See also SaaS or custom development.

Is AI-written code secure? Not guaranteed. According to Veracode 2025, 45% of test tasks ended with vulnerable code, and in 2026 the figure improved but remains a significant problem.

Who came up with vibe coding? Andrej Karpathy, in a post of 2 February 2025.

Sources

Want to check this on your store?

AI automation

An AI assistant checks the store every day, suggests fixes and makes them after your «Yes».

About the service →

Share

Related

  • Agentic commerce

    | Article: 5 min read

    Agentic commerce is shopping in which an AI agent acts on a person's behalf: it searches for and compares products, and within the permission it has been given it can also place the order and pay. For a store, this means part of the "visitors" will be…

  • Technical debt

    | Article: 6 min read

    Technical debt is the future cost of fixing and reworking that arises when software was built fast and without room for change. Like financial debt, it lets you move faster today, but with interest: every next change costs more and takes longer.

  • How an online store team uses AI every day

    | Article: 2 min read

    AI in a store team's daily work is a set of small, repeating tasks in which a language model prepares a draft and a person checks it and decides. The value comes not from separate "smart" features but from simple habits assigned to specific people.

  • AI automation for an online store

    | Article: 5 min read

    AI automation for an online store means handing part of the repetitive work to a language model (ChatGPT, Claude, Gemini and the like): draft descriptions and translations, a first reply to a customer, search through a knowledge base, spreadsheet analysis. It…