AI-Runnable Business Ontology: From Static Diagrams to Executable Knowledge Assets

Many enterprises produce artifacts such as data dictionaries, architecture diagrams, process models, and metric definitions that become obsolete after projects end because they are only human-readable and not machine-usable. The first edition of this tutorial showed how to structure business knowledge into an explicit ontology of concepts, relations, and rules.

With the rise of AI, the requirement has shifted: AI agents need structured, machine-interpretable knowledge to read documents, reason, and execute actions. A static diagram cannot be consumed by AI; without machine-readable structure, reasoning and execution are impossible, making governance ineffective.

The second edition centers on the concept of an AI-runnable business ontology , summarized in four Chinese characters: 读 (Read) , 推 (Reason) , 行 (Act) , and 治 (Govern) .

Read: The ontology serves as the AI's cognitive skeleton, providing a verified logical chain instead of letting the model guess.

Reason: Rules participate in inference; conclusions are derived from facts with traceable, explainable reasoning chains.

Act: Capabilities and preconditions are registered in the ontology, constraining agent actions within defined boundaries.

Govern: Changes propagate, versions are traceable, and accountability is assigned to organizational roles, ensuring the ontology evolves with the business.

The tutorial integrates established methodologies: Ontology Development 101 for scoping via competency questions; TOGAF and DEMO for a four-category meta-model skeleton; GraphRAG for human-machine collaborative construction; the evolution from RAG to Agentic RAG for retrieval; ontology-driven agents as a trustworthy foundation; and a CRISP-DM-style implementation process.

The author aims to solve three problems: (1) judge whether a project warrants building an ontology — not every project needs one, and scope must be defined first; (2) learn a step-by-step method from scoping and skeleton definition through human-machine collaborative construction to feeding the ontology to AI and driving agents; (3) learn governance to keep the ontology alive, aligned with business change, and demonstrably valuable.

All case studies are set in a manufacturing context — integrating MES, production scheduling, BOM extraction, equipment fault diagnosis, automatic ticket closure, and process knowledge retention — and are explicitly labeled as training examples not tied to real enterprises. The methods are claimed to be universally applicable.

The article concludes that no matter how advanced AI becomes, it requires a trustworthy semantic foundation, and that foundation is the ontology — now one that must be able to run.