Vibe coding: what it means if you're not a developer
In February 2025 the AI researcher Andrej Karpathy described a new way of making software: describe what you want, accept what the AI writes, and “forget that the code even exists”. He called it vibe coding. By November, Collins Dictionary had made it its Word of the Year.
For a business owner who is not a developer, it is a genuinely big change: you can now get working software from a plain-English description. It is also easy to overestimate. Here is where it helps, where it bites, and how to tell the difference.
Where it is genuinely good
- Prototypes. Before you pay anyone to build an app, you can build a rough version yourself in an afternoon and see whether the idea works. That is an enormous saving on ideas that turn out not to.
- Internal tools. A calculator for quotes, a tidier way to view a spreadsheet, a checklist app for your team. Small, private, low-stakes.
- One-off jobs. Converting a messy export into a clean list, renaming four hundred photos, merging two contact lists.
- Explaining what you want. A clickable prototype is a far better brief for a developer than a document. It cuts the misunderstanding that makes projects run over.
Where it bites
The trouble starts when a prototype quietly becomes the real thing — when it takes customers' details, handles payments, or runs part of the business every day.
- Security. Generated code often works while being unsafe: passwords stored carelessly, secret keys left visible in the page, databases readable by anyone who knows where to look. It is invisible until someone looks, and then it is a data breach.
- Destructive mistakes. In July 2025, an AI coding agent on the Replit platform deleted the live database of an app that SaaStr's founder was building, during what was supposed to be a code freeze — then claimed, wrongly, that the data could not be recovered. Agents with permission to change live systems can make large mistakes quickly.
- Maintenance. Software is not finished when it works. It needs updating when the things it depends on change. Code nobody understands is code nobody can safely change.
- Rules you did not know about. Data protection, accessibility, payment rules. The AI will not raise them unless you ask, and you cannot ask about a rule you have not heard of.
A simple rule for deciding
Ask one question: what is the worst thing that happens if this is wrong?
- “I waste an afternoon” — vibe code it yourself. Have fun.
- “A customer's data leaks”, “money goes to the wrong place”, “the business stops” — have a professional review it before it goes live, or build it properly.
A useful middle path: prototype it yourself, then pay a developer to review and harden it. That is often much cheaper than a from-scratch build, because the hardest part — working out exactly what you want — is already done.
A checklist before anything goes live
- Does it store personal data? Where, and who can read it?
- Are there any keys or passwords in the code the browser downloads?
- Is there a backup, and has anyone restored from it? (Backups are not backups until you restore.)
- Can the AI tool change or delete live data without asking you first?
- If the person who built it disappeared, could someone else maintain it?
What it means for what software costs
It is reasonable to expect small software projects to get cheaper, and they are. AI tools are part of how modern software is written, including ours. What they have not removed is the judgement — knowing what to build, what can go wrong, and what the law requires. That is where the cost of a professional now concentrates, and it is where you should expect to pay.
If you have built something and want to know whether it is safe to put in front of customers, send it to us. And if you are weighing up a bigger build, when custom software is worth it sets out the test we use.
Sources
Researched, written and published by Aftab Ahmad, owner of Seven Layers.