The room had twelve tables. Germany kept sitting at the wrong one.

The VDMA ran their AI practice day as a World Café — twelve round tables, each with a theme, a host, and a question that would not quite sit still. You rotated. You talked. You listened to people who build things for a living try to figure out what artificial intelligence is actually for.

The tables covered the range: AI in technical sales, governance, the EU AI Act (with Dr. Gorenflos fielding questions from companies who still are not sure which risk tier they live in), agents in production, agents in the back office connecting ERP and Office, edge AI, physical AI, sovereignty, strategy, culture, training. A full map of where the industry thinks the questions are.

Then there was the table the organisers called the "Wilde 13" — the break table, where the prepared agenda ran out and people said what they actually meant. The prompts there were better than any of the official ones. One in particular: *"AI strategy — not 'how fast can we adopt?', but where should it lead us — and how do we ensure AI strengthens our strategy, not replaces it?"* That question deserved the main stage.

Here is what I came away with.

**The optimisation reflex**

German mechanical engineering is very good at improving what it already has. That is not a flaw — it is the entire basis of the Mittelstand's quality advantage. You iterate. You tighten. You get 3 percent better every year and you compound. The trouble is that this reflex, applied to AI, produces process optimisation almost exclusively: take the workflow you have, add a model, run it a little faster. Comfortable. Incremental. It returns something measurable by Tuesday.

The same reflex is one reason China and the US are building capabilities Germany does not yet have. Not because their engineers are cleverer, but because they more readily accept that some things need to be built from scratch rather than improved from where they stand. I say this as someone made of German engineering, not as an outside critic. The reflex is earned. It is also, right now, a trap.

Process optimisation via AI is not wrong. It is just the wrong layer to optimise first.

**The layer worth building**

The honest question from the break table was about strategy — where does this lead? I think the answer starts lower than most companies expect: with data. Not "data strategy" in the consultant sense. With the genuinely unglamorous prior question: *what data do we have, and where does it come from?*

Transparency first. Before an agent can do anything useful, someone in the organisation needs to be able to answer that question without guessing. That is the foundation. It is not exciting. It is also not optional.

From there the sequence runs: a platform that makes that data reliably accessible (not a data lake where things go to be forgotten, but an API surface that applications and agents can trust); then a read-only agent that can make recommendations — a person still decides; then, once you have earned the right to trust the system, a predictive layer with a human in the loop, gated by audit trail.

You cannot optimise your way to a capability you do not yet have. A faster workflow built on an opaque data history is still opaque. The new foundation is not in conflict with the optimisation instinct. It is what makes optimisation valuable instead of merely busy.

**The room's own success factors**

The conference's stated success factors for production AI — context and grounding, reliability, fine-grained permissions, human-in-the-loop, traceability, guardrails — are not aspirational. They are a description of minimum viable trust. A person decides. Real numbers only. An audit trail. Those are not features we are planning to add. They are the posture we started from. The Wilde 13 question and the twelve-table agenda are, in that sense, the same conversation Apuna has been having with clients since the beginning.

The industry is asking exactly the right questions. The answers are already available. The missing piece is not insight — it is the will to build the foundation rather than optimise around its absence.

*The VDMA Praxistag KI im Maschinen- und Anlagenbau was held 18 June 2026 in Frankfurt am Main, organised by VDMA Software & Digitalisierung.*