Building Agentic AI Systems: From First Agent to MCP, Multi-Agent Orchestration, and Production

This practical guide takes readers from the principles behind AI agents to building and operating them for real customers. It begins with the agent loop, tool use, routing and structured outputs, then explores MCP, multi-agent collaboration, context and memory, retrieval, planning, durable workflows, browser agents and frameworks. Readers write the agent loop before using frameworks, and can run the Python examples with the Claude API or an open model on their own computer.

The book also addresses the challenges of putting agents into production: security risks such as prompt injection, approval controls and scoped identity; evaluation, tracing, reliability, load, latency and cost; and decisions about models, platforms and build-versus-buy. Its customer-focused engineering guidance follows the path from turning a broad request for AI into a measurable brief to piloting against agreed criteria and handing over a system the customer can own.

Across 32 chapters, six interludes and seven capstones, the material includes 249 exercises with worked solutions, reference resources and projects spanning customer support, data analysis, incident triage, code review, research and back-office workflows. A case study follows an agent through release, incident and rollback. Four reading paths accommodate beginners, agent developers, production engineers and architects.