AI-assisted engineering: AI agents do the groundwork, the engineer signs off
Tested software, a short shutdown window, a program your maintenance team can read and documentation that matches the final state. AI agents handle the inventory analysis, program drafts, testing in simulation and documentation; review, safety and commissioning stay with the responsible engineer.
- of schematics turned into an I/O list and tag table in about eleven minutes
- 51 pages
- decides, reviews and signs off
- One engineer
- on your project data
- No training
In short
For us, AI-assisted engineering means one responsible engineer plans and reviews the Siemens control project while a well-established set of AI agents takes on part of the routine work. The agents analyse schematics and existing programs, draft blocks, tags and screens against the functional specification, test the program and HMI in simulation and keep the documentation up to date. The engineer sets the structure and style, reviews every result, programs the safety functions himself and commissions the plant on site. You receive a deterministic program in TIA Portal, tested and documented, in standard formats. No AI runs in the controller.
Evidence: a 51-page schematic in about eleven minutes
From a 51-page schematic in PDF format, the AI agents produced an I/O list and, from it, the tag table for TIA Portal in about eleven minutes. Without them, this groundwork means typing everything up by hand, and those engineering hours are then missing from the plant logic.
- Result: an I/O list and tag table as the basis for programming or a retrofit.
- The engineer then checks the result against the drawings before it goes into the project.
- The time gained goes into the questions that need experience: what happens on a wire break? How does the plant restart after a power failure?
- We do not put a figure on the time saved across a whole project. That depends on the plant, the documents and the state of the existing code.
Where AI agents contribute to the project
The AI agents work only on the engineering side; on the plant itself, the engineer does the work. Within the project, they take work off his hands in four places:
- Inventory analysis: schematics, I/O lists and existing programs are analysed, including S7-300 STL code without symbols. The result is a functional specification of the current state. Every statement in it can be traced back to a point in the code or the documents; anything that cannot be proven is listed as an open question.
- Program drafts: recurring blocks, tags and WinCC Unified screens are built against the functional specification, as is the translation of existing code to the target platform. The engineer sets the program structure, programming style and tag naming.
- Testing: the program and HMI run together in simulation. Sequences, edge cases and faults are worked through; every tag has to arrive correctly on screen, every button in the controller. The tests can be repeated after every change, and the results are logged.
- Documentation: the functional specification, comments and test logs are produced alongside the project and updated with every change, including for existing plants whose documents have gone missing.
What stays with the engineer
Decisions and sign-offs are made by the engineer, David Prybisch, with more than 12 years of PLC programming and over 30 projects and commissioning assignments on four continents. He directs the AI agents and is responsible for:
- The functional specification, agreed with you. No programming starts without it, because it is the yardstick for the program, the tests and acceptance.
- Program structure, programming style, tag naming and screen concept. These rules are the framework the agents work within.
- A full review of the program and HMI. A first draft rarely fits straight away; corrections go back to the agents until both follow the rules.
- The safety functions. The engineer programs safety-related program parts himself and validates them the conventional way.
- On-site commissioning: downloading to the real controller, testing automatic mode, HMI and safety functions on the plant, and seeing it through to acceptance.
Your project data
A TIA project holds the process knowledge of your plant. Clear rules govern how we handle it:
- No training: your project data is never used to train AI models.
- Our own systems: we work on projects on our own systems rather than in public chat tools.
- Data flows agreed in advance: before the project starts, we agree with you which documents go into AI-assisted processing and which do not. Any requirements from your IT or data protection team belong in exactly this conversation.
- Confidentiality: on request, we sign a non-disclosure agreement (NDA) before the project begins.
What you receive
The outcome is an ordinary control program. No AI runs in the controller: the CPU executes a deterministic program that can be monitored online, changed and maintained like any other.
- A TIA project with meaningful tag names, network titles and comments. In migrations, the program is rebuilt in FBD or LAD, so your maintenance team can see in online status which condition is missing when a fault occurs.
- An HMI in WinCC Unified with a consistent screen concept, derived from the PLC structure.
- The functional specification, test logs from the simulation and the commissioning acceptance report.
- Everything stays with you, in standard formats. To maintain it, you need TIA Portal, not AI and not any tools of ours.
Four packages to get started
Clearly scoped services for a first step: we review your documents, then quote a fixed price. Larger projects work the same way, with the engineering at a fixed price after review and commissioning usually billed on a time-and-materials basis. The initial consultation is free of charge.
- Schematic inventory analysis: an I/O list and tag table from your PDF drawing set, as a basis for programming or a retrofit.
- Re-documentation: a functional specification derived from existing code, including S7-300 STL without symbols.
- WinCC Unified screen package: screens, alarms and tags derived from the PLC structure, tested in simulation.
- Retrofit check: effort estimate and hardware mapping for moving from S7-300 to S7-1200 or S7-1500.
Related services
- PLC Programming (TIA Portal)Structured programs from a proven block library — tested before commissioning.
- Retrofit S5/S7 → TIA PortalMigrate S5, S7-300 or S7-400 to TIA Portal on S7-1500/S7-1200 — often without rebuilding the control cabinet.
- Virtual CommissioningPLC program and HMI tested in simulation before the on-site date, every function backed by evidence in the test report.
Frequently Asked Questions
Who is responsible for AI-generated code?
The engineer. Responsibility for the delivered program lies with him, not with the tool, just as it would with a compiler, a block library or a code generator. Every block is reviewed and tested in simulation before delivery. Liability and warranty are set out in the quotation, as with any engineering contract.
How is quality checked?
Against the functional specification, which is agreed with you before the first line of code. The project is compiled, every block is compared with the functional specification, the program is tested in simulation and the HMI operation is checked against the simulated controller. The engineer reviews the program and HMI in full, and you receive the test results as a log. On the plant, commissioning follows through to the signed acceptance report.
What happens to our project data?
It is never used to train AI models. We work on our own systems rather than in public chat tools and agree with you before the project starts which documents go into AI-assisted processing. On request, we sign a non-disclosure agreement (NDA). If your IT or data protection team has requirements for external contractors, we go through them in that same conversation.
Does the AI also write safety programs?
No. The engineer programs safety-related program parts himself and validates them the conventional way; he tests the safety functions on the plant during commissioning.
Will AI be running in the controller afterwards?
No. The CPU executes a deterministic program in FBD, LAD or SCL that can be monitored online, changed and maintained like any other. The AI agents work on the engineering only, never on the plant. If plant data is to be analysed with AI later, that happens read-only via OPC UA and outside the controller, as described on the Plant Data & OPC UA page.
Can we maintain the program without AI?
Yes. You receive an ordinary TIA project with tag names, comments and a functional specification, in standard formats. Your maintenance team or another contractor can keep working on it in TIA Portal, without AI and without any tools of ours. In migrations, the program is rebuilt in FBD or LAD, so it stays readable without knowledge of a high-level language.
Is this AI consulting?
No. Prybisch Automation is an engineering office in Luxembourg for Siemens automation: we deliver control programs, HMIs, documentation and commissioning. AI-assisted engineering is how we produce those results. We do not offer consulting or training on introducing AI in your company. How AI is used in PLC programming today, and which questions to ask a contractor, is covered in our guide: AI in PLC programming.
Have your documents reviewed
Tell us about the plant, the controller and the documents you have, such as schematics as PDF, the existing program or a requirements specification. After the review, we suggest a sensible first step and quote a fixed price for the engineering.
✓ Non-binding initial consultation✓ Response time < 24h on business days