New 'Knowledge Architecture' paper redefines data engineering for AI era

A new paper titled "Knowledge-Centric Information Systems" by Mariano Garralda-Barrio proposes that the rise of large language models forces a fundamental shift from traditional data engineering to a new discipline: knowledge architecture. The author argues that organizational knowledge is no longer a passive informational resource but an executable infrastructure that systems retrieve, assemble, reason over, and act on. This requires redefining classical data-engineering guarantees as the unit of management shifts from records to knowledge artifacts.

The paper provides a detailed taxonomy of these redefinitions: ETL becomes knowledge ingestion, change-data capture becomes knowledge change detection, lineage becomes provenance, and catalogs become knowledge catalogs. Medallion architectures evolve into raw–curated–operational knowledge layers. Emerging formats like LLM Wiki and the Open Knowledge Format (OKF) are cited as early evidence of this transition. The central claim is that knowledge architecture becomes essential when organizational knowledge must be operationally delivered to humans, agents, workflows, and models to execute work.