“You’re Right to Be Afraid”
What a metalworking entrepreneur near Munich understood about AI that most consultants ignore.
By Sven Vollmer · The Industrial Translator
It was at the edge of an event, over a glass of wine, that the real conversation began. Not the one from the stage — the other one. The kind where politeness ends and the honest questions start.
Across from me stood an entrepreneur I’ve respected for years. Metal fabrication, roughly 200 employees, just outside Munich. The kind of company you’ll find hundreds of in this country — solid, grown over decades, no start-up shine, but full order books and a workforce of skilled tradespeople, many of them there for twenty years or more.
He got to the point quickly. “Sven, tell me honestly: what am I supposed to do with this AI?”
And before I could answer, the rest came out — all at once, like something that had been pressing on him for a while.
He kept hearing the same thing from every direction. He had to deploy agents. He should let AI run the shop floor. Replace customer service with chatbots. Every talk, every consultant, every trade journal carried the same message: automate, replace, cut headcount — or the competition will leave you behind.
“And you know what really worries me?” he said. “My son and I are standing in front of this decision together. And I have the feeling that if I get it wrong, I don’t just lose money. I lose my people. And with my people, I lose what they know.”
Then he looked at me and said the sentence this article is about:
“Maybe I’m just too old and afraid of change.”
I set down my glass and told him he was wrong. Not out of politeness. But because he was right about almost everything — except that last sentence.
The Fear Isn’t a Weakness. It’s an Early Warning System.
Over the past three years, we’ve rehearsed a strange casting of roles. Whoever is for AI counts as forward-thinking. Whoever hesitates counts as behind the times — someone who “slept through the change.”
That framing is convenient. And it’s wrong.
Because what my counterpart voiced that evening was not fear of technology. It was entrepreneurial judgment. Without using a single piece of jargon, he had named the single greatest risk of any AI rollout in a mid-sized industrial company: the loss of knowledge that no one ever wrote down.
At that moment, an experienced advisor should not reassure. He should confirm.
Because this man’s fear is not a weakness to be coached away. It is a sensor. A very good one. It points precisely to the spot where most industrial AI projects will later fail — the difference being that the others don’t know it yet.
He knew it. That’s why he was right to be afraid.
The Error Hides in a Single Word
Look at what he’d been advised to do. Let AI “run” the shop floor. “Replace” customer service.
Buried in those phrases is an assumption no one states out loud, because it’s treated as self-evident: that the purpose of AI is to remove the human. That success is measured by how many people are no longer needed at the end.
That’s the error. And it isn’t a technical flaw — it’s a flaw in thinking.
Whether AI replaces or supports is not a property of the technology. It’s a decision. A leadership decision. The same language model, the same agent can be deployed to make a person redundant — or to make that person stronger. The technology is identical in both cases. What differs is the question you ask of it.
“From every direction” he heard that he had to. But “have to” is not a technical category. No one has to let AI run the shop floor autonomously. You can decide for it — or against it. And that decision doesn’t belong in the IT budget. It belongs at the table where two people were sitting that evening.
Why “Replacing” Is Especially Expensive in Mid-Sized Industry
Here’s the point where I have to object as an engineer — not as an AI enthusiast.
An AI agent learns from data. Fine. But what, exactly, is the data it should learn from when the question is whether a part is “good enough,” which supplier actually delivers when things get tight, or why you lay this one weld seam differently for this particular alloy than the textbook says?
That data sits in no system. It sits in the head of the Meister — the master craftsman who, in German industry, is far more than a foreman: certified, responsible, the technical conscience of the shop floor.
The experiential knowledge of a tradesperson who has stood at the same line for twenty years is not a “soft factor.” It is the training data you cannot buy. It is precisely what a model would need in order to know what “right” even means — and it is written down nowhere.
Whoever replaces that person to save costs is not doing what he thinks. He believes he is cutting a task. In reality, he is destroying the only source from which an AI in his company could ever have learned. He is sawing off the branch the model is sitting on.
And it gets worse. I’ve written elsewhere about error propagation in agent chains: when one agent makes a mistake and the next accepts it as fact, the risk multiplies through the process until it surfaces, expensively, somewhere downstream. What is the most effective safeguard against this silent escalation?
An experienced human who looks at it and says: “That’s not right.”
The experienced employee is not a cost factor. He is the guardrail made of flesh. He is the instance that stops a hallucinating agent before a data error becomes a production stoppage. Whoever rationalizes him away is saving on the brakes to pay for the engine.
The One Question That Decides Everything
My counterpart didn’t need a technology seminar that evening. He needed a distinction he could hold on to. Here it is.
The wrong question is: “What task can AI take over?”
That question almost always leads to replacement — because almost anything can be answered with “in principle, yes.”
The right question is: “What task should AI amplify?”
And for that question, there is a surprisingly clear dividing line:
What is rule-based, repetitive, and documented may go to an agent. Reconciling order confirmations. Monitoring delivery dates. Pre-sorting standard customer inquiries. Here AI relieves the burden — and frees up exactly the people who have been misused as “data postmen.”
What rests on experience, judgment, and context stays with the human — amplified by AI, not replaced by it. The read on whether a supplier is a risk despite good numbers. The call on whether to make an exception for this customer. The feel for the machine that tells you something doesn’t sound the way it should.
The art isn’t in drawing the line once. It’s in drawing it deliberately — as a leadership decision, not as the prescription of a consultant who doesn’t know the shop.
The Meister’s Role Doesn’t Disappear. It Gets Promoted.
What’s needed here I’ve been calling Adult Supervision for two years now. The term is deliberately pointed. It says: this technology is too consequential to leave to itself. It needs people with process knowledge and judgment who decide what the agent may do, where it creates value, and where it does harm.
And this is exactly where the answer to the entrepreneur’s fear lies.
In a properly understood AI operation, his experienced craftsman does not become redundant. He gets promoted. The one who used to perform the routine task becomes the one who supervises the AI that performs it. His experiential knowledge shifts from silent background know-how to the most valuable resource in the company: he becomes the authority that judges whether the machine has it right.
That is not a threat to experiential knowledge. It is its promotion.
What I Told Him
I did not tell my counterpart that evening to stop worrying. That would have been wrong — and he would rightly not have believed me.
I told him something else.
Don’t start with the technology. Start with your people. Walk through your operation and, for every task, don’t ask: can AI take this over? Ask: is this routine — or is this experience? The first, you may automate. The second, you must protect and amplify. And when you’re unsure whether something is routine or experience, ask the person who’s been doing it for twenty years. He knows.
He thought for a moment, then said: “That I can explain to my son.”
That’s all it takes. No AI strategy with forty slides. One clear distinction that an entrepreneur can explain to his successor at the kitchen table.
His fear was not a sign that he’s too old for the change. It was a sign that he’d understood what matters — earlier than many of the people trying to sell him AI.
He was right to be afraid. And that’s exactly why he’ll get it right.
And you?
If you walked through your operation today — which task would you hand to an agent, and which would you never take out of the hands of your most experienced people? Drawing that line is not a technical exercise. It’s the most important leadership decision of the coming years.
E-Mail: sven.vollmer@business-quotient.com
Sven Vollmer is “The Industrial Translator.” He bridges the gap between industrial operational reality (SAP, supply chain) and the possibilities of generative AI. His focus is on value-creating applications—beyond the hype.
Transparency Note: This article was created with editorial support from AI (Gemini/Claude). The ideas, technical validation, use case selection, and adult supervision were 100% authored by Sven Vollmer.
LinkedIn: www.linkedin.com/in/sven-vollmer-bq
