Knowledge management is probably the best-funded and most disappointed discipline in the history of organizations. Three decades of document bases, intranets, enterprise wikis, communities of practice, capitalization portals — and the verdict, in nearly every organization, holds in two scenes: repositories no one consults, and experts one keeps calling because « he, he knows ».
The lazy diagnosis blames the users — no time, no discipline, no culture of sharing. The honest diagnosis is more disturbing: KM mistook its object. It archived what experts produce — documents — believing it was capturing what they know. But a document is a result without its reasoning: the conclusion without the discarded hypotheses, the recommendation without the signals that triggered it, the procedure without the thousand cases where it does not apply. Everything that makes an expert's value — the variables he watches, the patterns he recognizes, the errors he has learned to avoid — was never in the deliverables. KM filed shadows cast.
We built libraries of results believing we were preserving reasonings.
Three structural flaws followed. Knowledge without context: a document torn from its original situation — who, why, against which alternatives — becomes illegible to whoever was not there; hence the document graveyards. Charity as a business model: KM asked experts to document on top of their work, with no durable attribution or benefit of their own — the contribution was a tax, the evaporation a consequence. Search as the sole mode of access: one already had to know what to look for in order to find it — KM answered the questions one asked, never those one did not know to ask, though that is where the value sleeps.
Then generative AI arrived, and many believed the problem was vanishing: one would « chat with one's documents ». This is KM's error, accelerated: conversing with shadows cast produces fluent answers over an amputated body of knowledge — and unsourced ones at that. The fluency of access does not correct the poverty of the object.
The rupture is not to index documents better. It is to change object: to preserve reasoning — attributed, contextualized, with the consent and under the name of its bearer — and to structure it so that it meets the problems it can solve, including those no one thought to submit to it. This is no longer document management. It is another category.
That other category has a name: cognitive preservation — reasoning preserved under its name, discovery rather than search. See Reveal and Cortex →