A sentence broke off at a dinner table in Stuttgart a few weeks ago. I had just described what I am building with Pheraxia — deliberately lean, no classic departments, headcount kept minimal by design. Someone across the table said, “But then you’re not even…”
The sentence never finished. We both understood what it meant.
I have heard variations of it in my AI empowerment workshops across the DACH region for months. The resistance rarely starts as a question about tools or models. It starts as a question about identity. And until you address that question honestly, no AI implementation survives first contact with your organisation.
The Deepest Resistance to AI Is Not Technical
The operating logic of most successful Mittelstand companies was built around a simple equation: more responsibility means more people means more status. The CFO with a department of twelve carries more weight than the one with a team of three. The manager who built a division earns more trust than the one who has not yet.
AI disrupts that equation directly. With the right tools and proper coaching, a senior expert can produce outputs that previously required a team. And the first thing many seasoned people feel when they truly understand this is not excitement. It is the anxiety behind that unfinished sentence in Stuttgart.
Then I’m not even…
I am not raising this to be philosophical. I am raising it because organisations that skip this conversation will deploy AI for the wrong reasons, to the wrong people, and then wonder why adoption stalls six months in.
We Give AI to the Wrong People
The instinct in most rooms is to hand AI to the youngest, most digitally comfortable person available. Let them figure it out and train the rest later.
I have come to believe this is exactly backwards — and consistently damaging.
Here is the structural problem with the junior-first model: AI produces output that looks highly competent. Fluent sentences, structured analysis, neatly sourced arguments. Reviewing that output well requires knowing, almost instinctively, when something is subtly wrong. That is not a skill you develop in two years. A talented analyst early in their career will miss what a seasoned operator catches in thirty seconds — not because the analyst is less capable, but because they have not yet built the pattern recognition that only comes from years inside a specific market, a specific customer base, a specific regulatory environment.
In workshops where senior experts received structured coaching — not just a licence and a login, but deliberate work on real tasks — I observed a significant gap emerge over a matter of weeks. Coached, experienced practitioners consistently outperformed untrained use of the same tools. The experience did not become obsolete. It became the multiplier.
Put your most experienced people on AI first. Not because they are the most comfortable with new technology. Because they are the ones who know when the output is wrong.
The Three Costs of AI
This month on the 10xCEO podcast, Christoph Maichel and I ran a small but instructive test. We gave identical research briefs to two AI systems. Both returned roughly thirty sourced statements each. Both missed the same thing: a privacy clause published only days earlier. A third system, assigned only to check the first two outputs, found it.
That experiment clarified something I keep returning to in conversations with operators and owners across the region.
The Three Costs of AI — a framework:
| Layer | Cost | Who Owns It |
|---|---|---|
| Generate | Near zero | Anyone with a licence |
| Verify | Significant | Requires lived context and expertise |
| Trust | Highest | Cannot be replicated in two weeks |
Generating content with AI costs almost nothing today. Every competitor in your market can do it. Verifying that the output is actually correct — that it reflects your customer relationships, your regulatory context, your company’s real position — requires judgment that takes years to develop. And the trust a client places in your conclusions, your contracts, your signed advice, is something no tool can manufacture on demand.
The competitive moat in an AI-saturated market is not generation. It is the verification and trust layers. Both of those live inside your senior people.
Think of every AI output as a first draft from a brilliant but inexperienced intern — someone with seven degrees and zero days inside your company. Impressive range, genuine effort, occasionally and confidently wrong. The expert reviews it. The expert signs it. That is the only quality gate that holds.
Same Licence, Different Chance
There is a second dimension to this that deserves more space in Mittelstand leadership conversations: equal access is not the same as equal opportunity to build competence.
A LinkedIn News DACH feature this month on the Gender AI Gap surfaced a data point from Initiative D21 that struck me as an organisational problem more than a capability problem. They measured a sixteen percentage point gap in AI usage between men and women. Where employers paid for structured training, that gap narrowed to one percentage point.
The same pattern shows up in teams I work with, regardless of demographics. A licence distributed to the whole organisation does not produce equal learning. The people who get protected time — two structured hours per week, on a real task, inside work hours — build genuine competence. The people who are expected to learn in the margins of an already overloaded schedule do not.
Distributing access without distributing time is not an AI strategy. It is a way of preserving the advantages that already existed before the tool arrived.
What to Do Next Week
Here are five things you can act on before Friday:
-
Audit your current AI deployment. Who is actually using the tools consistently? Are they the people with the deepest contextual knowledge of your business — or the most available ones?
-
Protect learning time for your seniors. If your most experienced people are exploring AI in the margins of overloaded schedules, adoption will stall. Block two hours per week, on a real work task, not a sandbox demo.
-
Build your verification layer before you scale. Before you automate any workflow, define clearly who checks the output and on what criteria. If the answer is “no one yet,” stop and build that layer first.
-
Have the status conversation explicitly. In your next leadership offsite, ask directly: if headcount is no longer the primary measure of scope and impact, what replaces it? The organisations that answer this question thoughtfully will retain their best people through the transition. The ones that avoid it will lose them to confusion.
-
Name your trust moat. Ask your team: what is the specific thing a competitor cannot replicate in two weeks, even with the same AI stack? That answer is where your strategy should focus — and where your senior people belong.
That broken sentence in Stuttgart stayed with me because I recognised the feeling behind it. I am building Pheraxia from zero — no prior experience in biotech, no established playbook, minimal team by design. Every day I am the Level 2 intern who needs the domain expert in the room to tell me I am wrong. That dependency is not a weakness. It is the quality gate that keeps me from being led confidently in the wrong direction.
AI amplifies whoever holds it. The question worth asking is not whether your organisation is using AI. It is whether the people who know what good looks like are the ones deciding what ships.
Who in your leadership team has had protected time to build real AI competence — and who is still trying to learn in the margins?