AI FOR PLC CODE
AI Copilots for PLC Programming in 2026: Siemens, Rockwell, CODESYS and Beckhoff
Every major PLC platform now has an AI assistant. Here is what each one can generate, where they fall short, and a step-by-step way to verify AI-written ladder and Structured Text.
In 2026, every major PLC platform has an AI assistant. Siemens' Engineering Copilot TIA (in preview) generates SCL, test cases and WinCC Unified screens; Rockwell's FactoryTalk Design Studio Copilot writes ladder and Structured Text routines in its cloud IDE; CODESYS sells an MCP server that lets AI agents edit Structured Text in a project; and Beckhoff has announced TwinCAT 3 CoAgent for Engineering. They draft and explain code well, but none removes the need to review, simulate and test what they write.
Checked 1 October 2026. Features and status come from each vendor's own product pages and documentation, listed under Sources. These products change quickly, so check the vendor page before you buy.
Which AI copilots exist for PLC programming?
Siemens: Engineering Copilot TIA
At CES 2026 in January, Siemens announced nine new AI copilots across design, lifecycle management, manufacturing and operations, including ones for Teamcenter, Polarion and Opcenter, offered through the Siemens Xcelerator Marketplace. For PLC engineers the relevant one is Engineering Copilot TIA, which works alongside TIA Portal.
Siemens' product page for Engineering Copilot TIA Standard (Managed Service), still marked as a preview, lists:
- generating SCL code from descriptions, creating test cases for SCL blocks and documenting code;
- ladder (LAD) generation, marked "coming soon";
- creating and configuring WinCC Unified HMI devices and screens, JavaScript help and VBScript migration;
- configuring SIMOTICS motors and encoders for drives;
- translating project texts, explaining project context and answering TIA Portal questions.
Licensing is per seat: one seat can be assigned to up to three users but used by one at a time, with 600 chat requests and 2,000 uploaded document pages per seat per month. Siemens provides the AI services and cloud infrastructure.
Rockwell Automation: FactoryTalk Design Studio Copilot
FactoryTalk Design Studio is Rockwell's cloud-based engineering environment for Logix controllers. Its Copilot is a generative AI assistant that can answer questions from the help, release notes and error documentation; generate PLC code; create Add-On Instructions, smart objects, programs and routines (ladder diagram or Structured Text); find references in a project; and explain features and errors. Searching the project and generating code need a Core entitlement.
Rockwell's documentation is refreshingly direct about limits. It states the Copilot may display inaccurate information, that it can explain but cannot modify safety-related PLC code and device configuration, and that you must send an update command after editing the project by hand so the Copilot sees the current version. Conversations are private to the user and not used to train Rockwell's or Azure OpenAI's models.
You may also see Rockwell's LogixAI mentioned. It is not a coding assistant: it is a ControlLogix chassis module, launched in 2019, that learns from controller data to detect production anomalies and act as a virtual sensor.
CODESYS: the MCP Server
CODESYS took a different route. Instead of building its own chatbot, it sells the CODESYS Development System MCP Server, an add-on that lets any AI assistant supporting the Model Context Protocol work directly inside a CODESYS project: reading, writing and editing Structured Text, navigating the project, managing libraries and reading compile errors and warnings. Two limits matter. It supports Structured Text only (objects in graphical languages can be read but not created or changed), and CODESYS warns that with a cloud model, all project data the user can access is sent to the provider. Because CODESYS V3 has limited undo, CODESYS recommends file-based project storage with Git. In its internal tests, CODESYS found Claude models gave the best Structured Text.
Beckhoff: TwinCAT 3 CoAgent
Beckhoff showed TwinCAT 3 CoAgent for Engineering (TE1700) at Hannover Messe 2026. Beckhoff describes PLC code suggestions from plain-language requirements and project context, I/O configuration suggestions, HMI layouts from sketches and direct access to the Beckhoff Information System, with support for several cloud and local language models and MCP. When we checked, the product page still said "product announcement", with market release on request. A companion product, TF1700 CoAgent for Operations, targets operation and diagnostics.
General assistants
Many engineers simply paste a description into a general assistant such as ChatGPT, Claude or GitHub Copilot and copy the Structured Text back. That works for small blocks, but the assistant knows nothing about your project, your tag names or your vendor's dialect unless you tell it, and you are responsible for what you paste in.
What can AI copilots do well, and what can't they do?
Good at:
- boilerplate: function block skeletons, state machine frames, alarm handling patterns;
- explaining inherited code you did not write;
- documenting code and generating first test cases;
- translating project texts and scaffolding HMI screens;
- converting between similar dialects, with checking.
Not good at, or not allowed to do:
- knowing your plant: what each input really means, mechanical timing, which interlocks matter;
- guaranteeing safe behaviour on faults, stops and power loss;
- safety code (Rockwell's copilot is explicitly blocked from modifying it, and that is a sensible rule for every tool);
- scan-cycle subtleties, first-scan behaviour and retained data;
- staying inside one vendor's dialect: drafts can mix TIA Portal SCL and CODESYS habits, or call instructions and library blocks that do not exist.
Research supports the caution. In the LLM4PLC study (ICSE 2024), researchers put grammar checking, compilation and formal model checking into a loop around models including GPT-4. Generation success rose from 47% to 72% and expert-rated quality from 2.25 to 7.75 out of 10. Even with automatic verification, roughly one program in four still failed.
How do you verify AI-generated ladder or SCL?
Treat AI output like code from a capable new hire who has never seen the machine: probably mostly right, never trusted unchecked.
A worked example: the tank fill that looks right
Ask an assistant for "SCL to fill a tank: open the inlet valve on start, close it at high level, alarm if it takes too long". A typical draft opens the valve on a start pulse, closes it when the high-level switch is true, and starts a TON timer inside the IF branch that runs while the valve is open. It compiles. Now review it against the checklist:
- The timer is only called inside the IF. In TIA Portal and CODESYS, a TON instance that is not called every scan does not see its input go false, so it may not reset and the next fill can alarm at once. Call the timer every scan and drive its IN input from the condition.
- No safe state. What happens if the high-level switch fails? Nothing closes the valve. Add a second, independent high-high cut-off or at least a maximum fill time that closes the valve, not only raises an alarm.
- No restart rule. After a power cycle, does the valve reopen by itself? Decide, then code it: most plants want a deliberate restart.
- Stop is missing. The prompt never mentioned a stop, so the draft has none. AI writes what you ask for, not what the machine needs.
None of these are exotic; they are the faults experienced PLC engineers check for by habit. The only way to catch them in AI code is to have that habit yourself.
Simulate before anything goes near hardware: PLCSIM for TIA Portal, Logix Emulate or Logix Echo for Rockwell, and CODESYS Control Win for CODESYS. Drive the inputs through normal sequences and every fault you can think of. Our comparison of free PLC simulators shows what each one can do.
Why do fundamentals still matter?
Because the copilot drafts and you sign off. A reviewer who does not understand the scan cycle, timers, edge detection, interlocks and safe states cannot see what is missing, and missing logic is the most common AI error. The engineers who get the most from these tools are the ones who could have written the code themselves, and use AI to write it faster and document it better.
There is also the people who maintain the machine. A technician fault-finding at night needs clear, conventional ladder or SCL with sensible names and comments, not clever generated code nobody on site understands.
How to learn it
- AI Coding for Engineers: Claude Code, Cursor and Copilot covers writing a spec before the prompt, reviewing every change, using Git and tests, and generating Structured Text that is checked and simulated. It includes an introduction to Siemens' Industrial Copilot for engineering.
- Siemens SCL Structured Text in TIA Portal gives you the SCL you need to review a copilot's output.
- CODESYS and IEC 61131-3 Programming teaches portable Structured Text in a free environment.
- Allen-Bradley Studio 5000 covers Logix ladder and tags for the Rockwell side.
To try AI drafting safely, the free Structured Text Generator produces an IEC 61131-3 draft for CODESYS, TIA Portal SCL or Studio 5000, and the PLC Code Explainer turns code into plain English with a list of risks. Both produce drafts and explanations to check, not code to download to a machine. For the bigger picture, read PLC programming languages in 2026 and generative AI for engineers, or start from free PLC training.
The courses are free in full. The optional EDWartens Certificate of Completion starts from US$2.99 and is not a Siemens, Rockwell, CODESYS or Beckhoff certification.
Take the free course
Questions
Can AI write PLC code?
Yes, it can draft it. Siemens, Rockwell, CODESYS and Beckhoff all offer or have announced AI assistants that generate Structured Text, SCL or ladder from a description. The drafts are useful starting points, but research and vendor documentation both show they can be wrong, so every line must be reviewed, simulated and tested before it runs a machine.
What is Siemens Engineering Copilot TIA?
A generative AI assistant for TIA Portal. Siemens' preview of Engineering Copilot TIA Standard generates SCL from descriptions, creates test cases and documentation, builds WinCC Unified screens, helps configure SIMOTICS drives and answers TIA Portal questions. Ladder generation is listed as coming soon. It is licensed per seat with monthly limits on chat requests.
Can Rockwell's copilot write ladder logic?
Yes. FactoryTalk Design Studio Copilot can create ladder diagram or Structured Text routines, Add-On Instructions and programs from a prompt, and explain existing code. Code generation needs a Core entitlement. Rockwell's documentation warns it may display inaccurate information and states it can explain, but not modify, safety-related PLC code.
Is it safe to paste my PLC project into a cloud AI?
Check your company's rules first. CODESYS warns that with a cloud model, all project data the user can access is sent to the provider's servers. Rockwell says Design Studio Copilot conversations are not used to train its or Azure OpenAI's models. Never paste credentials, IP addresses or customer data into a public chatbot.
Do I still need to learn PLC programming if AI can write it?
Yes. You can only review code you understand, and responsibility for what the machine does stays with the engineer who signs it off. Scan cycles, timers, interlocks, safe states and fault handling are exactly where AI drafts go wrong, and maintenance technicians still fault-find in ladder at 3 a.m. without an assistant.
Sources
- Siemens: Siemens unveils technologies to accelerate the industrial AI revolution at CES 2026
- Siemens: Engineering Copilot TIA Standard (preview)
- Rockwell Automation docs: FactoryTalk Design Studio Copilot
- CODESYS Development System MCP Server documentation
- Beckhoff: TE1700 TwinCAT 3 CoAgent for Engineering
- Beckhoff: TwinCAT projects with AI-supported engineering
- Automation World: Rockwell Automation LogixAI module (2019)
- Fakih et al., LLM4PLC: verifiable programming of PLCs with LLMs (ICSE-SEIP 2024)
Written by the EDWartens engineering team for general education. Product names are trademarks of their owners; mentioning them does not imply endorsement. Prices and terms of other providers were checked on the date shown and can change.
