
Graph Engineering for Agentic AI Systems: Design the topology. Don’t inherit it.
Author(s): Ken Huang (Author)
- Publisher Finelybook 出版社: Packt Publishing
- Publication Date 出版日期: October 29, 2026
- Edition 版本: 1st
- Language 语言: English
- Print length 页数: 372 pages
- ISBN-10: 1809208270
- ISBN-13: 9781809208279
Book Description
Design reliable agentic AI systems with explicit graph topology, controlled multi-agent orchestration, knowledge graph memory, and verifiable execution paths
Key Features
- Choose agent orchestration patterns that justify their complexity and cost
- Apply graph design through worked examples, chapter exercises, and companion code
- Test workflow paths, recovery behavior, and approval gates before agents take action
Book Description
Agentic AI systems need more than capable models: they need clear rules for how work runs. Graph Engineering for Agentic AI Systems helps you design that structure, starting with when a single-agent loop is enough and when multi-agent orchestration earns its complexity. Distinguish execution graphs, knowledge graphs, and learned graphs, then explore typed state, routing, parallel fan-out and joins, and human approval gates. Compare runtimes including LangGraph, Google ADK, Microsoft Agent Framework, and the OpenAI Agents SDK.
Worked examples, chapter exercises, and companion code connect architecture to implementation. Explore knowledge graph memory and GraphRAG, checkpointing, bounded retries, idempotency, and production observability. Examine permissions and authorization boundaries, then evaluate execution traces, node contracts, and stop conditions. By the end, you’ll be equipped to design, test, and operate agent workflows with explicit topology, controlled side effects, and traceable decisions.
What you will learn
- Choose when to keep an agent loop or introduce a graph
- Design typed state, routing, parallel fan-out, and joins
- Compare agent runtimes for orchestration and persistence
- Build knowledge graph memory with typed, traceable facts
- Operate workflows with checkpoints, retries, and idempotency
- Verify execution paths, permissions, and human approval gates
Who this book is for
This book is for AI engineers, LLM application developers, software architects, and platform engineers building agent workflows beyond the prototype stage. Security engineers will also benefit from its treatment of agent identity, authorization, and approval boundaries. Familiarity with Python, LLM applications, and software architecture will help you engage with the code examples and exercises.
Table of Contents
- Graph Engineering Is Topology You Own
- Execution Graphs and Memory Graphs
- The Graph Under the Harness
- Anatomy of an Agent Graph
- When Not to Build a Graph
- Patterns That Earn Their Nodes
- Frameworks You Can Ship an Agent Graph On
- Knowledge Graphs as Memory
- Operating Graphs in Production
- The Computer Science of Agent Graphs
- Security, Identity, and Governance of Agent Graphs
- Graph Intelligence Is Not the Runtime
- Testing, Evaluation, and Verification of Agent Graphs
- Keep the Graph After the Name Fades
Editorial Reviews
Editorial Reviews
About the Author
Ken Huang is an author and researcher in AI applications and agentic AI security, serving as CEO and Chief AI Officer at DistributedApps. He is Co-Chair of AI safety groups at the Cloud Security Alliance and the OWASP AIVSS Project, and Co-Chair of the AI STR Working Group at the World Digital Technology Academy. He is an EC-Council instructor and Adjunct Professor at the University of San Francisco, teaching GenAI and agentic AI security to data scientists. He co-authored OWASP’s Top 10 for LLM Applications and contributes to the NIST Generative AI Public Working Group. His books are published by Springer, Cambridge, Wiley, Packt, and China Machine Press. A frequent global speaker, he participates in major technology and policy forums.
finelybook
