January 26, 2026 · aiindustrial-automationvirtual-commissioningsiemenstia-portal
Something is shifting in industrial automation, and most people in the field have not noticed yet.
AI is entering the control system development workflow. Not as a replacement for engineers, but as a tool that fundamentally changes how fast and how reliably we can deliver projects. I have been experimenting with this for the past year, and the results have surprised even me.
The Tools Are Getting Interesting
Modern PLC development platforms are starting to look less like traditional automation tools and more like software development environments. Siemens’ Simatic AX, for example, brings PLC programming into VS Code with text-based Structured Text. That matters because text-based code is something AI models can actually work with.
But the feature that changed my thinking is AxUnit testing.
AxUnit lets you write unit tests for your function blocks and run them without needing PLCSim. Not even a simulated PLC. Just run the command and get results in seconds.
For anyone who has waited for PLCSim to start up just to test a small logic change, you know why this matters.
Interface-Based Simulation
Thanks to OOP support in Simatic AX, you can define interfaces for your hardware dependencies. Instead of your controller talking directly to a real sensor, it talks to an interface. Then you can swap in a simulated sensor that returns whatever values you need.
Say I want to test an oven temperature controller. I define an ITemperatureSensor interface. My controller uses that interface. For testing, I inject a thermal model simulation that calculates temperature based on heater output — kind of like the physics would actually work.
Now I can run many test cycles in seconds. Test sensor failures. Test edge cases that would be dangerous on real equipment. I can test my function blocks to about 80% readiness before even opening PLCSim.
This is dependency injection applied to PLC programming. It is the same pattern software developers have used for decades, and it is finally practical in our field.
Real-World Result: 5 Days Instead of 3 Weeks
Theory is interesting, but results are what matter. Here is what happened on a recent project.
I was commissioned to build the control system for an industrial oven. This kind of project — PLC programming, HMI development, I/O configuration, commissioning — normally takes about 3 weeks from arrival on site to a running system.
I knew I was going to use my AI-assisted workflow, so I quoted 2 weeks.
It took 5 days.
I should say — I did not do it alone. I built a team of AI agents that handle the routine design work. The functional structure design. The function block programs. The system simulation model and test blocks.
The system model let me run virtual commissioning and catch most mistakes before I even arrived at the customer site. That part was huge, actually. By the time I walked into the building, most of the logic had already been tested against a simulated version of the oven. The on-site work was mostly verification and fine-tuning, not debugging from scratch.
The Workflow: AI Executes, Engineer Reviews
Here is how the workflow actually works in practice:
- I define the requirements. What the machine needs to do, the I/O list, the control philosophy, the safety requirements.
- AI agents handle the scaffolding. They generate the functional structure, write standard function blocks, create documentation, build test models.
- I review everything. Every function block, every connection, every safety-related piece of logic. The AI handles volume. I handle judgment.
- Virtual commissioning catches errors early. The simulation model runs test cycles and flags issues before real hardware is involved.
- On-site commissioning is faster. Because most logic errors were caught in simulation, the on-site work focuses on hardware verification and final tuning.
They execute. I review and refine.
What AI Can and Cannot Do Today
I want to be honest about where this stands, because there is too much hype in this space.
What AI handles well:
- Generating standard function blocks from specifications
- Creating documentation and wiring schedules
- Building simulation models for virtual commissioning
- Scaffolding project structures and boilerplate code
- Repetitive design tasks that follow established patterns
What AI cannot do (yet):
- Make engineering judgment calls about safety
- Design novel control strategies for unusual processes
- Replace the experience of watching a machine run and knowing something sounds wrong
- Handle the unexpected problems that only show up on site with real hardware
- Understand the full context of a customer’s process requirements
The key principle: every line of code is verified by an experienced engineer. The AI accelerates the work. It does not replace the expertise needed to validate it.
Why This Matters for Project Timelines
The industrial automation industry has a well-known problem: projects take too long and cost too much. Commissioning timelines slip. Engineers are scarce. Customers wait months for systems that should take weeks.
AI-assisted development does not solve all of these problems. But it compresses the parts of the work that are repetitive and time-consuming — the parts that used to eat up days of an engineer’s time doing things a structured process can handle.
The result is not just faster delivery. It is more reliable delivery. When you catch errors in simulation before arriving on site, you avoid the costly cycle of debug-fix-download-test that stretches commissioning timelines.
The Shift
I used to be a PLC programmer. Now I am becoming a manager of digital employees. Still getting used to that idea.
The tools are ready. The workflow is proven. And for companies where commissioning delays cost real money, this changes the equation.
Planning a project? Contact us to discuss how we can help.
