Your intern built a dashboard with AI, so why pay a development team? Because AI-generated code degrades fast, skips the safety checks that protect your business, and quietly turns a quick win into an expensive problem.

Someone on your team (maybe an intern, maybe a junior analyst) just built a working dashboard using Lovable, ChatGPT or Claude. It looks good. It works. And now you're sitting in a meeting asking a very reasonable question:

‍Why am I paying a software company when my people can do this themselves?

It's a fair question. AI code generation is genuinely impressive. Tools like Claude, Cursor and Copilot can produce software in minutes that would've taken days a year ago. We use these tools ourselves at Kohde; they're part of the modern development toolkit now.

But there's a meaningful gap between "it works" and "it's ready for production." And that gap is where businesses get hurt.

Vibe Coding Risks: The code works, until you change one thing

Here's what actually happens when you build with AI and skip professional development practices.

"I've got a little app I'm busy writing," says Joe van der Walt, Kohde's Founding Director. "I tell the AI to build a feature, and it builds it. Fast. Then I come back the next day and say, just change this one small thing. It changes it, and then it doesn't work anymore. I say fix it; it still doesn't work. So I open the code to see where it's getting stuck, and it's a complete mess."

AI writes "plausible" code. It's optimised to generate something that "plausibly" works . But it doesn't maintain code, and maintenance is where the real cost of software lives. 

That first feature might take a day or two. But the small change afterwards? That can take a week to untangle, because the AI has built layer upon layer of tangled logic with no consistent structure underneath.

This isn't just Joe's experience. A 2025 study by CodeRabbit, analysing hundreds of open-source projects, found that AI co-authored code contained almost double the major issues as human-written code. The code compiles and runs, but it's fragile in ways you won't see until something breaks.

It's not just code, it's everything around the code

When people think about software development, they picture someone writing code. But at a professional level, writing code is only one part of the job.

"There's a big part of our job that's just writing code," Joe explains. "But there's also a lot around it: the way you write code, how you review it across a large team, the way of working. You could generate the code and not save a lot of time, because you still need to do all the associated thinking."

That "stuff around the code" includes architecture decisions that determine whether a system can scale, code reviews where another developer checks your work before it goes live, automated testing that catches bugs before users do, and deployment pipelines that move software safely from development to production. 

These aren't glamorous, but they're what separates software that works for a demo, from software that works for a business. 

The dangerous part? AI skips all of it.

The risk you can't see

This is where it gets serious for business owners. When a non-developer publishes AI-generated code, there's no one checking what that code actually does under the surface.

"As soon as you publish that thing on the internet," Joe warns, "you need a person in the middle to check what it actually did. Because it can leak client information, and you won't know it, because you're not a software developer.."

This isn't theoretical. In early 2026, a social networking platform built entirely through vibe coding suffered a major data breach. Security researchers found that the AI had configured the database with wide-open public access, exposing over a million authentication tokens and tens of thousands of email addresses. The founder had never reviewed the infrastructure code.

For South African businesses handling client data (particularly in finance, healthcare or any regulated industry), this kind of exposure isn't just embarrassing. It's a compliance risk with real consequences.

The real cost: Multiply it across your business

One broken feature that spirals into a week of fixes is frustrating. But the real damage shows up when you scale this pattern.

"If you extrapolate that over three projects," Joe says, "that's a lot of money you're going to waste."

Vibe coding front-loads speed AND back-loads cost. The dashboard your intern built in an afternoon might work fine as a proof of concept. But the moment you need to change it, integrate it with another system, or hand it to someone else to maintain, you're paying for all the shortcuts at once. 

A professional development team might take slightly longer upfront, but the total cost of ownership is significantly lower because the foundation is built to last.

AI is a tool, not a team

None of this means AI is bad for software. At Kohde, we use AI tools daily; they're excellent at accelerating the mechanical and repetitive parts of development. The difference is that our developers direct the AI, review every line of output, and maintain the discipline and quality processes that keep software stable over time.

The question isn't whether AI can write code, it’s who's accountable when something goes wrong?

If you're weighing up whether to build something with AI tools in-house or bring in a professional team, we're happy to have that conversation. No pressure, just an honest look at what makes sense for your situation.