Multi-agent systems represent a new stage in the development of applications based on generative artificial intelligence. But how should they be designed and coordinated properly? In what cases should they be applied? In this practice-oriented book, Victor Dibia shares his experience in creating multi-agent tools such as AutoGen, AutoGen Studio, and Agent Framework, helping the reader understand the principles of these systems, their areas of application, and approaches to their architecture, development, and implementation.
Book Contents
The book covers 15 chapters, contains more than 185 code snippets, and is illustrated with 46 original diagrams. It addresses basic concepts and interaction patterns of agents, step-by-step creation of systems from scratch (including the development of PicoAgents), methods for evaluating and optimizing performance, as well as real-world application examples—from data analysis to software development and information processing.
Approaches to Learning
Special attention is given to learning from first principles: instead of studying a specific framework, which may become outdated, the reader consistently creates their own PicoAgents library, including agents, tools, memory, workflows, and orchestration mechanisms for multi-agent systems.
