Every executive faces the same calculation today, and it seems to have no good answer. On one side, a clock: in expertise-heavy sectors, an entire generation is leaving, carrying off know-how no document retains. On the other, a homogenization: when every organization in a sector runs the same model, all end up producing the same thinking. Urgency pushes toward adopting AI as fast as possible; adopting it without precaution erases the very thing that sets you apart.
The two apparent options
First option: deploy generative AI as it comes. It can invent a reference without anyone noticing, with the same assurance as an established fact — and a single false claim in a signed deliverable is enough to dent a reputation built over thirty years. It gradually aligns your reasoning on a statistical average, and what made people come to you, specifically, fades. And it routes your grey matter toward infrastructure you do not control: what your teams deposit there today will feed someone else's model tomorrow.
Second option: stay on the sidelines. But generative AI has revealed a productivity gain your competitors are already exploiting — and with each week without adopting it, the distance grows. Meanwhile, your experts leave without a trace and their reasoning evaporates. Abstaining is not neutral: it is a cost, and it accumulates.
Adopt or abstain: both options cost you your singularity — by dilution on one side, by falling behind on the other. When two opposite outcomes lead to the same loss, the choice is wrongly framed.
What the market offers — and what is missing
Look at the available answers. Storage spaces file your documents but know nothing of their contents. Generic assistants converse, and invent. Enterprise search engines retrieve a file, not a line of reasoning. Heavy analytics platforms demand armies of integrators. Each solves a piece. None answers the question that matters: how to trust what AI asserts, while keeping control of what your organization knows.
What is missing is not one more tool in an already crowded stack. It is a foundation — something that sits beneath the tools, invisible, and guarantees what none of them guarantees: provenance, verifiability, sovereignty.
Mapping the dilemma
The dilemma is false, and to see it you need only map it on two axes. The first: is your knowledge extracted — poured into a model that feeds on it — or preserved, structured and attributed at your side? The second: is your reasoning capability dependent on a third party who can cut off access, or sovereign?
Two axes, four positions. Three are untenable over time — one is not.
Three squares fill in at once. Generic assistants occupy the extraction-dependence space: your reasoning trains someone else's model. Sovereign LLMs move up a notch — European servers, control of hosting — but remain extractive by design: sovereign in location, not in paradigm. And there is a seductive, fragile square: preserving knowledge without guaranteeing its sovereignty. It does not hold — carefully structured knowledge hosted by a third party remains at the mercy of a decision that is not yours. June 12, 2026 reminded the whole market: a frontier model made inaccessible overnight by a foreign state's decision. Preservation without sovereignty: a fine house on land you do not own.
Two crowded quadrants. One unstable. One occupant.
That leaves the fourth square — sovereign and preserving — and it is the only durable one. Knowledge there is structured, attributed, kept at its owner; reasoning there is verifiable, auditable, replayable years later, whatever model has passed in between. It is not one more position on the map: it is the only one that does not collapse when pushed through time. It is the one (Urs) occupies.
The third way is not a compromise
Between deploying and abstaining, one would expect a middle path — a bit of both, cautiously. There is none. The third way does not split the difference: it changes the axis. It shifts the problem: not should we adopt AI, but on what conditions — and those conditions, sovereignty and preservation combined, define a position no one else occupies. The executive who reaches it does not choose between relevance and singularity. They keep both.