Agentic GraphRAG

Part I. Foundations of Graph-Based Agentic Systems

Part I establishes the foundational theory of the book and introduces the end-to-end example you will build for the rest of it. Before we can construct an agentic system, we have to be honest about why the current generation of agents keeps stalling out in production: stateless retrieval, flat embeddings, no temporal model, monolithic prompts, and ad hoc tool integration. Those five flaws are not separate bugs. They are symptoms of a single representational gap, and closing that gap is what this book is about.

From there, we introduce the structural answer: the move from strings to things, the dual-graph architecture of a vertical knowledge graph (what the agent knows) and a horizontal workflow graph (how the agent acts), and the eight pillars that turn that structure into a working system. By the end of Part I, you will have a complete mental model of where we are going and why, along with the autonomous DevOps agent as a concrete system to anchor every later chapter.

• Chapter 1 defines agentic AI along three dimensions of agency, places systems on the workflow-agent spectrum, diagnoses the five flaws of vector retrieval-augmented generation (RAG) for agents, and motivates GraphRAG as the evolution of contextual retrieval.

• Chapter 2 introduces the dual-graph architecture, lays out the eight pillars and how they map to the rest of the book, and introduces the autonomous DevOps agent in its initial reactive state.