Knowledge as Infrastructure: How AI Reveals Knowledge Management as a Systems Discipline

Vontava

Most organizations deploying AI have treated knowledge management as a background dependency, something that feeds the model, not something that has to be designed for it. My latest paper challenges that assumption directly.

When a human reads a knowledge article, they evaluate it. They notice when it's outdated, incomplete, or contradictory. When an AI system consumes that same article as retrieval context, a training document, or a prompt component, that evaluation layer disappears. Knowledge that was once reference material becomes load-bearing infrastructure, and the engineering discipline required to maintain it changes accordingly.

The central argument is this: AI does not create a problem with knowledge management. It reveals that knowledge management has always been a systems discipline. The scope adapts. Nothing is negated.

The paper works through what that adaptation looks like in practice. It introduces knowledge interfaces as the architectural unit of analysis, maps how governance requirements differ depending on how AI uses knowledge downstream, and addresses the organizational question most teams are currently avoiding: who owns this? The answer is that the knowledge team and the AI team own different parts of a shared system, and neither can do their job without the other. Eight principles anchor the practical guidance.

Three companion reference papers go deeper for architects who want them: a maturity model, a worked example in a RAG-based service desk deployment, and a reference architecture showing how the full system fits together.

This paper builds on the progressive enrichment model I introduced in "Data in the Moment" (2025) and extends the architectural framing from my ServiceNow Knowledge 2026 session.

If you work at the intersection of knowledge management, enterprise architecture, or AI deployment, this was written for you. I'd love to hear what resonates.