PLC programmers learn their craft in automation, electrical or mechanical engineering. They know machines, processes and commissioning. Software engineering is rarely part of that foundation.
What gets skipped in daily project work
A concept before the first function block, an architecture, a test concept, version and document management: in software development, these are standard. In machine building they often get skipped, not out of carelessness, but because time and method are missing. The consequences show up later, at commissioning, at the next update, or when a colleague takes over the program.
What an AI assistant takes over
An AI assistant can take over much of that work: drafting concepts, developing and documenting test cases, writing documentation, checking code for vulnerabilities, managing versions. Above all, it keeps to a fixed process, from requirements through architecture, test and documentation to release, even when the deadline is tight.
What’s involved
- Model: your personal assistant. It guides you through the process and is available around the clock.
- Harness: the assistant’s working environment. It connects the assistant to your development tools; this is where the guardrails take effect.
- Rules: fixed working instructions per project and a standardized process. They tell the assistant how to work.
- Skills: packaged domain knowledge and workflows the assistant loads when needed.
- Agents: helpers with their own narrow task, for example a reviewer that checks documents.
- Vault: a linked knowledge base, the memory across projects.
- Four-eyes principle: a second model reviews the first model’s result before you accept it. It complements compiling, static analysis and tests; it does not replace them.
- Guardrails: limits the assistant cannot get around: permissions in the working environment block what it may not do, and automatic checks such as compiling, static analysis and tests verify the result. Rules guide, guardrails enforce. They are not optional.
- Agentic coding: the assistant works through a task on its own, in several steps. At the end of the course, you build such an agent: after a program change, it creates test scenarios for it, runs them together with all existing tests and documents the result. You review the new scenarios before they join the test suite; changes to the machine code stay with you.
Why now
AI assistants are changing how software is built, in automation too. Your experience stays at the core: knowledge of the machine, the process and commissioning comes from you, not from the AI. Learning to lead the assistant makes that knowledge more effective and lets you shape the change.
From practice
This way of working grew out of DDAS’s own development of CODESYS-based machine software and libraries. We are turning it into an online course: self-paced, shown on an example machine, with a CODESYS-specific package of process agent, skills and vault as added value.
Questions and topic requests to contact@ddas.digital.