AI at work in the Infracom web department

For us, AI is no longer a hype, but a practical tool that helps us make our workflows smarter and faster. We actively use AI for development, carefully preparing our instructions and prompts. The AI agent writes parts of our code, but always within our standards, ensuring the output matches what we would create ourselves. Everything AI produces is thoroughly checked and validated. This way, we maintain control over quality.

AI does more than just generate code

We also use it for documentation and, for example, to let AI create a decision tree for customers, allowing them to immediately verify whether changes have been implemented correctly. We also build proofs of concept with the help of AI, enabling us to quickly develop an interactive sample application for customers to try out—something we previously had little time for. AI also helps us verify our own time estimates: by presenting the assignment, elaboration and documentation to the AI, we can immediately see whether we overlooked anything.

For different projects, we use various AI models, such as Gemini, GPT and Claude. This allows us to use the best‑suited model for each challenge.

Taking over AI‑generated software

We sometimes take over AI‑generated software from customers. In those cases, it is important that human knowledge remains present. AI can do a lot, but it does not understand the underlying logic of a system. Software created by AI is only as good (or bad) as the input provided by humans. A solid functional design remains the foundation of every application, and human expertise is essential to guide, adjust and verify. AI executes; we keep thinking.

Vibecoding: the new hype, but not without risks

More and more people are discovering vibe coding, a way of developing where you tell the AI what you want to build using simple prompts, and the AI generates the code. Ideal for those who want to quickly develop a prototype without deep technical knowledge. In theory, anyone can have an app built simply by describing what they want. The big advantage is speed: an idea is turned into a working prototype in minutes. This makes technology accessible to everyone, offers creative freedom and saves costs.

Disadvantages of vibecoding

The AI makes choices you might not make yourself, and small details are difficult to control. Security and privacy are not always guaranteed, and those who cannot prompt well receive generic, sometimes unusable code. Many people also become dependent on platforms such as GitHub Copilot and Replit.

The biggest risk? Those who rely solely on vibe coding often lack fundamental system understanding. That means: if something goes wrong, you don’t know where to start or how to fix it.

No blind trust!

AI can help you tremendously, but it remains crucial to keep thinking for yourself, checking and understanding what you use. At the web and software department of Infracom, we ensure that AI is a tool—not something we blindly trust. That makes us not only faster, but also smarter and safer.