“Why are we still programming PLCs like it’s 1995?” Bernhard Böhrer, founder of logiccloud, poses this provocative question at the beginning of his presentation at the Stutensee editorial office’s trade press days. And he hits a nerve with it. Anyone who thinks back to the 1990s remembers AWL (instruction list), manual memory addressing and rigid hardware couplings. But while the IT world has reinvented itself several times over the last three decades, classic “hard work” still often dominates PLC development, despite the fact that virtual control systems have now been introduced.

The complexity of modern automation solutions is growing exponentially. Cloud connections, IoT interfaces, cybersecurity and more and more software variants are driving up costs. However, engineering budgets and, above all, the available skilled workers are only growing linearly, if at all. Productivity is becoming a bottleneck. Engineers spend too much time writing boilerplate code and standard routines instead of concentrating on process optimization. Or as Böhrer puts it: “Highly qualified engineers do busywork.”

The way out: the PLC becomes software

The approach taken by companies such as logiccloud breaks with two paradigms of traditional automation: the link to proprietary hardware and the local development environment. The controller becomes an “app”. Detached from dedicated PLC hardware, the runtime runs on any industrial PC or edge gateway, for example in cooperation with hardware partners such as InHub.

The entire development environment is also moved to the browser and is therefore accessible via the (German or private) cloud. Local installation is no longer necessary. Only this step into the cloud creates the data basis to ignite the next evolutionary leap: the use of generative artificial intelligence (GenAI) in the engineering process.

Architecture: First the plan, then the code

Many developers already use tools such as ChatGPT or other coding agents to generate code snippets. However, this does not go far enough for industrial automation. Generic AI lacks an understanding of the IEC 61131-3 standards, the hardware restrictions and, above all, the specific project context.

“A plausible code is not automatically a secure code,” Böhrer warns against blind faith in standard LLMs. In order to drive the hallucination out of AI, logiccloud relies on a specialized architecture that goes far beyond simple “prompting”. The “Project Assistant” is based on a pyramid of internal expertise, a technical knowledge base and AI models built on top of it. The workflow is similar to the way a human engineer works:

• 1. input & domain knowledge: The process begins with a project description in natural language (text) or a Q&A dialog.

• 2. planning engine: Before a line of code is written, a planning instance breaks down the requirement into logical sub-areas. The AI decides which artifacts are required: The actual PLC program, reusable components, I/O definitions and, a decisive difference to pure code generators, also the appropriate HMI screens.

• 3. validation: This is the critical gatekeeper. The generated output is not accepted unchecked, but validated against logiccloud rules and IEC standards. This is the only way to prevent syntactically correct but functionally nonsensical or unsafe code from entering the runtime.

Four modes of interaction: from “vibe coding” to reverse engineering

The AI is not a one-way street, but is integrated into the logiccloud IDE. The chatbot does not act in isolation, but knows the context of the entire project, including variables and architecture. “The AI works in the context of the project and not in a vacuum,” emphasizes Böhrer. In practice, four use cases can be distinguished:

• Code analysis and optimization (“What’s wrong?”): The AI analyzes existing routines for errors, inefficiencies or violations of best practices and suggests optimizations.

• Vibe Coding (“Add this feature”): A term that describes fluid AI programming. The developer gives rough directions, such as “add a fill level monitor”, and the AI adds the necessary code almost in real time, matching the style of the rest of the project.

• Logic reverse engineering (“What does this do?”): Particularly valuable for legacy projects or when familiarizing yourself with third-party code. The AI analyzes complex components and explains their function in understandable language.

• Tailored help (“How to do this?”): Context-related assistance with the implementation of specific tasks, replacing the need to wade through manuals.

The network effect of the cloud

Data storage is a technical aspect that is often underestimated. In traditional automation, terabytes of valuable project knowledge and solution strategies lie dormant on local hard disks of laptops, isolated in silos. This data is inaccessible for the training of AI models.

Thanks to the cloud approach, logiccloud centralizes the projects. This enables the models to be trained on an ever broader basis of real automation patterns, of course with data protection and consent. The AI learns with each project (“network effect”), recognizes patterns in the hardware connection or in standard control loops and thus becomes increasingly precise. This distinguishes the system from local on-prem solutions, which are, however, just as possible.

The Pareto principle: from coder to validator

The aim is not to replace the programmer, but to relieve them massively. The Pareto principle applies: around 80 percent of the code, including standard functions, I/O mappings and basic structures, is handled by the AI assistant. The remaining 20 percent, i.e. fine-tuning, safety checks and complex process logic, remains in the hands of the engineer. According to the speaker, the division is clear: “The AI takes care of the mechanical aspects, the human takes care of the meaningful aspects.”

This workflow changes his role: he has to type less, but check more. The validation of AI results becomes a core competence. At the same time, AI is solving another vexing issue in the industry: documentation. “No programmer I know likes documentation. Most of them just want to program,” Böhrer knows from practical experience. Using the reverse engineering functions, the assistant can analyze existing legacy code and automatically add comprehensible comments.

Outlook: Democratization of automation

In the long term, AI could significantly lower the barriers to entry into automation. The vision is that in future, process engineers or heating engineers without in-depth PLC programming knowledge will also be able to describe their requirements in prose and obtain a functioning basic application.

In addition, the AI should not only write software in future, but also make hardware recommendations. Based on the complexity of the code and the required cycle times, the system could suggest which control hardware is most economical for the specific project.

It is the step from manual manufacturing to industrialized software creation. In other words: the final update for the year 1995.