A Reference Architecture for Enterprise Agentic AI Infrastructure Platforms: Design Principles, Operational Lifecycle, and Production Deployment Patterns

Enterprise adoption of Agentic AI is accelerating as organizations integrate large language models, intelligent agents, retrieval-augmented generation, and autonomous workflows into business operations. While recent advances have significantly improved AI capabilities, there remains limited guidance for designing production-ready enterprise platforms that integrate these technologies with cloud-native infrastructure, governance, security, observability, and operational lifecycle management. This paper presents a vendor-neutral Enterprise Agentic AI Infrastructure Platform Reference Architecture that addresses these challenges through a layered architectural approach. The proposed architecture organizes enterprise AI capabilities into distinct Experience, Agent, Orchestration, AI Services, Enterprise Knowledge, Platform Engineering, and cloud-native Infrastructure layers while incorporating cross-cutting enterprise services, including security, identity, governance, compliance, and observability. In addition to the reference architecture, this work introduces complementary operational models covering the enterprise AI request processing flow, Day–0, Day–1, and Day–2 operational lifecycle, deployment models spanning on-premises, cloud, hybrid, and edge environments, cross-cutting enterprise services, and enterprise capability mapping. A qualitative evaluation demonstrates that the proposed architecture satisfies key enterprise quality attributes, including scalability, modularity, extensibility, portability, maintainability, interoperability, governance, and operational readiness. The resulting framework provides organizations with a structured, technology-agnostic foundation for designing, deploying, governing, and evolving production-scale Enterprise Agentic AI platforms while remaining adaptable to future advances in artificial intelligence and cloud-native computing. This work contributes a vendor-neutral enterprise reference architecture, an operational lifecycle model spanning Day–0 through Day–2 operations, deployment and governance frameworks for production environments, and a qualitative evaluation of the proposed architecture against key enterprise quality attributes.