The AI Copilot May Not Replace Control Engineers. Something Quieter Will.

Every other week now, someone on LinkedIn or Reddit announces that control engineers are next on the chopping block. Siemens has 120,000 engineers using its Industrial Copilot. Schneider's PLC Code Generation Copilot is being sold with claims of 30 to 50 per cent reduction in development time. Geoffrey Hinton tells CNN that 2026 will bring "a new wave of AI job losses." Pick a forum and you'll find an Instrument and Control System engineer asking whether they should change careers before it's too late.

I've been working in industrial control and safety systems for over a decade from simplest green field projects to most complex brown field migrations. I have an honest opinion about this, and it isn't the one being sold on either side.

What the copilots actually do

The Siemens and Schneider copilots are real tools doing real work. I will not disregard the role of AI in ICSS development and on plant maintenance. If you've spent an afternoon writing repetitive Structured Text for a pump start-stop logic, a copilot will save you time. If you need to generate boilerplate IEC 61131-3 from a natural language description, the current generation handles it reasonably well, particularly when the platform has been trained on the vendor's own libraries, but I would also like to point out that all these accessories so far implemented existed before in form of template example programs for each OEMs product line in relatively lesser interactive format.

That's the good case. The honest case is narrower than the marketing.

These tools generate code patterns. Code patterns are not the entire job. The job — if you've actually done it — is roughly thirty per cent code, fifty per cent design and coordination, and twenty per cent figuring out why something that should work doesn't. The copilot is helping with the first thirty.

Try this. Ask a copilot to write you a cause-and-effect matrix for a three-phase separator with high level, high-high level, and low-low level interlocks, where the SIS needs to coordinate with the BPCS without violating IEC 61511 separation requirements, the operator has requested that one specific shutdown valve fails open instead of closed for maintainability reasons, and the previous P&ID revision had an undocumented interlock that production wants preserved.

Now do it again with the actual tag numbers of your field instruments, a contractor's lead engineer who's two time zones away, and a client site manager who wants the FAT moved up by ten days.

You see that Control System projects are scenario oriented and every project is a story written from scratch.

What the panic gets wrong

Most of the people writing about AI in industrial automation have never been on a live plant. They've read the press releases. They imagine the job is "writing code." It' Not! Writing code is the visible artifact of the work. The work is making decisions, in the presence of incomplete information and incompatible stakeholders, that will hold up the next time the plant trips at three in the morning.

A copilot can suggest ladder logic. It cannot tell you that the level transmitter on T-401 has been giving spurious highs ever since they reused the impulse line from the old vessel, and that the operator workaround is to ignore alarms below 0.4 metres for the first eight hours after start-up. That information lives in the head of one fitter who's been there since 2009 and is retiring next March. The AI doesn't know it. Neither does the next graduate someone will hire to "use the copilot."

This is the gap the marketing skips over. AI tools are good at the part of the job that was already easy. They are weak at the part of the job that was always hard.

What might actually go wrong

Here's where I stop sounding reassuring.

The thing that worries me about AI in control systems isn't that it will replace experienced ICSS engineers. It's that it will quietly stop us from growing new ones.

When I came up, you wrote your first hundred logic blocks by hand. You made every mistake. You watched senior engineers reviewing the program, then explain why some rungs could go wrong. This trial, test, error and adjust loop is where experience is cultivated. By the time you were trusted with a real safety function, you had realized the patterns and really went through the drill of "What's wrong". You knew what good looked like because as you spent two years in repeating this loop on several projects while working in lesser responsible role.

A junior engineer in 2026 starts by typing a prompt into Industrial Copilot. The output is reasonable. They submit it. It passes review and the code compiles. They move on.

They never learn why it works. They never learn what would have been wrong. And five years later, when something on a live plant trips for a reason the copilot hasn't seen before, they neither have the foundation nor time to figure it out as things can get really tough in a tight commissioning schedule. That's just the logic part, set aside a whole different debate of hardware design troubleshoot in case something goes wrong.

This is the actual risk. Not job replacement. Skill replacement, one quiet generation at a time.

What I'd tell a junior engineer right now

If you're a junior I&C or control engineer reading this and wondering whether to panic: don't. The job isn't disappearing. It's bifurcating.

The engineers who treat AI as a faster way to skip the fundamentals will be unemployable in ten years. They'll produce code they can't explain, defend, or fix. The engineers who treat AI as a research assistant — useful for boilerplate, useless for judgement — will be more valuable than ever, because not everyone loves the drill. It takes time and patience to realize the concepts your learnt in your degree in a fully immersed Industrial environment.

A few practical suggestions, since you asked.

Spend the time AI saves you on the things it's not good at. Read the FDS line by line. Walk the plant with the senior operator. Ask the question that makes you look stupid. Learn what the field instrument actually does, not just what tag number it has.

When the copilot gives you code, regenerate the design problem in your own head before accepting it. If you can't explain why each rung is there, you don't understand the logic — and one day, you'll need to.

Specialise into the corners AI handles worst. Functional safety. SIL verification. Cybersecurity for OT. Brownfield projects where nothing matches the drawing. Anywhere accountability matters, you matter. It's the cornerstone of well executed ICSS Projects.

The thing nobody says

The truth I rarely see in these articles is that experienced control engineers are not particularly worried. We've watched three or four "this will change everything" cycles already. We watched DCS replace pneumatic. We watched the IT/OT convergence panic. There's still at least over thousands of facilities which still operate on pnematic control despite the rush of digitalization for two decades. We watched smart instrumentation, then digital twins, then Industry 4.0.

Every one of those was sold as the end of the discipline. None of them were. Each one shifted what the job looked like. The fundamentals — understand the process, design for failure, document everything, never assume — are still the fundamentals. Human element remains as relevant as ever in the age of AI.

Yes AI is another shift but it is not the end. The engineers who will be replaced are the ones who were already replaceable, doing work that AI is good at because the work was wrote to begin with. If that describes you, the answer isn't to be afraid of AI. It's to get better at the part of the job that requires you.

Or, put more bluntly: the copilot isn't going to replace you. It might replace the version of you that wasn't doing very much in the first place.

If you're working through any of this in your own career, I'd be interested to hear how it's landing for you. Reply or email me — I read everything that comes in.

© 2026. All rights reserved.

About Us

FAQ