An important development that we are seeing now? Vibe coding.
More and more, non-developers are using tools such as Cursor or Windsurf to develop software themselves. Not by writing code, but by talking to AI. They express their wishes in plain language; the AI generates the code. No more struggling with complex syntax or logic. The barrier between an idea and a working product has suddenly become very small.
For us, this is of course exciting. Part of what we, as software developers, did for our customers for years is now suddenly done by customers themselves.
Nevertheless, we have decided to fully embrace this development. Our customers are already working on it and I fully agree with them.
Vibecoding can be a great way to test ideas and automate your processes. At the same time, there are also limits to what you can achieve with vibecoding. And if you exceed those limits, there are risks.
That's what this article is about.
The 4 Levels of Software Development with AI
Where is the limit of what you can achieve with vibecoding? To make that clear, we distinguish 4 levels of software development with AI:
Level 1: Prototyping
The first level is prototyping. This means that you can quickly make your idea visual and tangible. You enter some prompts, the AI generates an interface and you can show stakeholders what you mean. A kind of digital sketch. As far as we are concerned, everyone can enjoy experimenting with this.
Level 2: single-user applications
At the second level, we talk about the single-user application. This is software that you actually build for your own use. Think of a handy tool that automates your specific workflow or a dashboard that only you can consult. It runs somewhere on a server or in a local environment and it works. Vibecoding is perfect for this type of application. The risks are relatively low and it can make your work a lot easier. Risks are limited to dealing with confidential information and data, but business continuity is not directly at stake.
Level 3: collaborative applications
Here we leave the safe haven. Multiple users log in. They work in the same database. They perform actions simultaneously. Suddenly, you have to deal with roles, rights and data integrity. This is where most encrypted applications start to falter. AI is great at writing functions, but it has trouble with the deep architecture of shared systems. Do you have some technical knowledge? Then maybe you'll figure it out. Do you miss that knowledge? Then it gets difficult. While the organization will become more dependent on the availability, continuity and proper functioning of the application.
Level 4: mission critical applications
At this level, you will find business-critical software that must remain running 24/7. Which many users should be able to trade at the same time. Where sensitive data is processed. It must be able to talk to other applications in your IT landscape. And what the continuity of your organization depends on. Safety, scalability and robustness are extremely important here.
For this level, we say: don't do it yourself. Call in an experienced engineer for this.
The hard line: why you shouldn't vibecode past level 3 (yet)
Up to level three, if you have the right digital skills and good analytical skills, you can go a long way. It's tempting to keep pushing afterwards. Why not also go to level four? But trust me: at the moment, that is extremely risky. The step towards mission critical is not linear growth. It's a leap into another universe.
The big risk of fully AI-generated software without human control over the architecture? The application may seem to work when you test it. But you have no idea what's happening at the back. Without you knowing it, there can be major security and scalability issues in the code.
That's why you should never start level four if you're not a senior engineer. The risks to the continuity of your company are simply too great. At this level, assurance becomes essential. You need specialists who understand how a system behaves under heavy loads and who see the security holes that an AI overlooks.
Two questions are crucial: what is your technical knowledge and what is your risk appetite? The higher the level at which you work, the heavier those two questions weigh. And the more security you need.
Hand over the baton to us and bring your app to the finish line after all
We believe in cooperation. You have an idea? Grab those AI tools. Build those prototypes. Bring your vision to life at levels one, two, or even three. We welcome that. It is the best briefing we could ask for.
Our service portfolio has been adapted to this. In levels 1 and 2, we help you come up with the right solution that will create real value, serve users well and achieve your business goals. Towards levels 3 and 4, our role becomes more technical: we guarantee security, performance, scalability and continuity.
By the way, that does not mean that we will rebuild everything by hand from phase 3 to phase 4. We have translated our many years of experience in software architecture into our own AI framework. We take your application and run it through that machine.
Our engineers use the AI framework to rebuild or optimize your application, based on our rules, guardrails and intelligence. This is how we make your application secure, scalable and future-proof in an efficient way.
Have you reached the limit?
Vibecoding is a powerful tool. Above all, use it. Experiment until you can't continue. But know when the time has come to call the professionals.
Software is the engine of your business. Make sure that engine not only runs on a good 'vibe', but on a strong foundation.
Tomorrow's winners are those who dare to pioneer now, but also those who understand where the limit of their own abilities lies.
Have you reached the limit? Then we will be happy to help you
