Photo: ©Zeugwerk GmbH/Raum15, D. Jelemensky
Many companies in mechanical and plant engineering are testing AI for PLC code right now. And are disappointed.
At first, the results look impressive. The AI produces code seemingly without effort. But in day-to-day use it quickly becomes clear that those results are barely usable in practice.
Why is that? Not because the models are poor, but because of a fundamental misunderstanding about how AI can actually work in machine building.
Most of the time the expectation is the same: give AI the right task description and it will write code. A few well-crafted prompts and suddenly you are more productive.
In practice, a very different picture emerges. Results are not reproducible. The code does not fit your own standards. Existing projects are not understood. Trust does not develop, and that is frustrating.
For developers it is tedious. For companies it gets expensive fast. Because suddenly you are spending more time checking, explaining, and reworking than you ever intended to gain. A productivity promise turns into additional overhead in the project day.
The problem is not the AI. It is the environment it is supposed to work in.
PLC development is not a creative free space. It is shaped by clear structures, defined architectures, and deterministic sequences. If AI is going to work here, it does not need better wording or more context in text form. It needs a clean, technical foundation to operate within.
The decisive question is therefore not how you use AI, but whether your own engineering system is built in a way that allows AI to work reliably inside it.
In my upcoming posts I will show why most approaches fail at exactly this point and what underlying structure is actually needed for AI to make a real difference in PLC development.
Because one thing is certain: AI will quickly become a permanent part of engineering. What is worth thinking about is the foundation it works on.
That is exactly where Zeugwerk comes in. Talk to us.