Agentic AI systems are rapidly being deployed as autonomous, goal-directed entities to manage the orchestration of complex, multi-step workflows across diverse domains. Despite their growing adoption, current frameworks often lack a formalized model and architecture. Hence, many implementations remain ad-hoc, relying on simplistic data structures and monolithic designs that hinder scalability, reusability, and interoperability. This paper addresses these limitations by introducing AgentO, an OWL/RDF-based ontology and accompanying knowledge graph that formally represent the core concepts, components, and interactions that underpin agentic AI workflows. Our ontology provides a standardized vocabulary for modeling agentic patterns including agents, tasks, workflows, and resource dependencies. To build and evaluate AgentO, we developed an automated LLM-driven process and translated 66 agentic workflows from four different agentic AI frameworks. We further evaluated our approach through three real-world use cases: declarative reconstruction of agentic patterns, cross-context reuse of tasks and agents, and agentic AI workflow auditing. Our results demonstrate the potential of semantic technologies to bring structure, reusability, and transparency to agentic AI systems.