“I’ll just build it myself with AI.” Plenty of business owners are saying it, and for a lot of small jobs it works. That habit has a name, vibe-coding, and the trouble starts when something built on a quiet Sunday ends up holding other people’s details.
This guide answers what vibe-coding is in plain words, why owners like it, what it tends to miss, and three checks anyone can ask for before launch. It also looks at a case MyBroadband reported on 16 October 2026, because it shows the gap clearly. We use AI every day at hatchd., so this is not a warning against the tools. It is about what happens after the build.
What is vibe-coding?
Vibe-coding means building software by describing what you want to an AI, accepting what it writes and running it, without necessarily reading the code. Andrej Karpathy, an AI researcher and a co-founder of OpenAI, put the name on it in a post on X on 2 February 2025. He described giving in to the flow, accepting changes without reading them, and pasting error messages back when something broke. He also framed it as good enough for throwaway weekend projects.
A vibe-coder, then, is anyone who builds that way. They are not always developers. Today it is just as likely to be a founder, a marketing manager or an operations lead with an idea and an afternoon. The “vibe” is the test: if it looks right and clicks through, it is done.
The loop, in three steps
- 1Describe itType what you want in plain words. “Make me a dashboard for my team.”
- 2Accept itThe AI writes the code. You take it as it comes.
- 3Run itIf it looks right, it is done. If it breaks, paste the error back in.
Why business owners like it
The appeal is real. It is quick: a first version in hours, not weeks. It is cheap: no developer queue to wait in. And it is useful, particularly for trying an idea, building a small internal tool or testing a page before you spend real money on it.
We use AI across our own work every day, and we would not tell anyone to stop. If you are testing an idea that holds nobody’s personal details, vibe-coding is a fine way to start. The risk comes from a different question: what happens when the quick build becomes something staff or customers rely on?
I use AI every single day, and I love it. It has changed how much I can get done. But I also know it gets a lot of things wrong, and it gets them wrong with complete confidence. So I always fact check it. Every number, every claim, every line it writes for something that matters. If I will not trust it unchecked in my own work, I would not trust it unchecked with your customers’ details either.
Working is not the same as safe
When nobody reads the code, three things tend to go unchecked. Nobody reads the code itself. Nobody tests who can reach the app. And nobody maps which real names, emails and numbers end up inside it. An app can do its job perfectly and still hand over more than you meant it to.
There is early evidence that AI-written code carries more problems on average. CodeRabbit, a company that sells automated code review, analysed 470 open-source pull requests and reported in December 2025 that the AI-authored ones averaged 10.83 issues each against 6.45 for human-authored ones, roughly 1.7 times as many. Two cautions apply. It is a vendor’s analysis of a modest sample, and CodeRabbit itself says it cannot be certain every pull request labelled human was written entirely by a person. Read it as a direction, not a verdict.
The Bootlegger BOS dashboard
On 16 October 2026, MyBroadband reported on a web application belonging to the coffee franchise Bootlegger, which it described as vibe-coded. The app, an internal operations dashboard called BOS, showed a login screen. But it sent a large amount of data along with the login page, so anyone with the address could read it in the page source. MyBroadband verified names and contact details linked to 132 Bootlegger locations, plus some performance and financial data. An anonymous poster said it started with staff photos of laptop screens on Instagram, which showed the site, and someone then looked it up.
The company responded quickly. CEO Ricky Ruthenberg said Bootlegger became aware on the afternoon of Thursday 15 October 2026 and took the legacy version of BOS fully offline the same day. He said BOS had already been moved to a secure environment, but historical data on the old platform had not been fully decommissioned. The company says no customer data was collected on the platform, that there is no evidence its databases or core systems were accessed, and that it will notify the Information Regulator under POPIA and contact affected franchisees, operators and managers. MyBroadband did not independently verify the customer data claim.
We are not saying vibe-coding caused this, and a developer typing every line can make the same mistake. The lesson is narrower and more useful: a login hides the page, but it does not hide what the page sends. And the gap here sat before launch, in an old app still online with old data, not in the response afterwards.
Three checks anyone can ask for
You do not need to read code to make the build safer. You need to ask the right questions, and make sure someone with the skills answers them.
Check three
- 1What goes in?List the names, numbers and anything private the app will hold. If the list is long, it needs a proper review.
- 2Who can reach it?Open it logged out, on another phone. Then ask a developer what the page sends before anyone logs in.
- 3Who reviews it?A person who reads code should look before strangers do. Then switch off any old version you stopped using.
In South Africa, personal information is covered by POPIA, which is why Bootlegger says it will notify the Information Regulator. A short review before launch costs far less than that conversation.
Use the tool, check what it sends
A sales system is built on your customers’ details: who enquired, what they asked and when you replied. That is exactly the kind of data that should never be an afterthought. We use AI to build faster, and we still check what goes in, who can reach it and who reviews it. If you want to see where an enquiry goes today, the 15-minute test takes one morning. And for the other side of how people respond to what you publish, see why likes measure other likes.
Sources: MyBroadband, CEO of top South African coffee franchise responds after vibe-coded web application exposed sensitive data (16 October 2026), and the earlier MyBroadband report. CodeRabbit, State of AI vs Human Code Generation, as reported by The Register (17 December 2025). Andrej Karpathy’s 2 February 2025 post on X, summarised at Wikipedia. Company statements are Bootlegger’s own, as reported by MyBroadband.