Systems Thinking for Agentic AI: Design reliable LLM and agentic AI systems with RAG, MCP, guardrails, evaluation, and observability
暫譯: Agentic AI 的系統思維：運用 RAG、MCP、護欄、評估與可觀測性設計可靠的 LLM 與 Agentic AI 系統

Apply systems thinking to build production-ready LLM and agentic AI systems using RAG, MCP, tools, memory, guardrails, evaluation, observability, and practical architecture patterns

Key Features:

- Architect reliable AI agent systems beyond prototypes and demos

- Build with RAG, MCP, tool calling, memory, and agent orchestration

- Engineer for safety, evaluation, observability, cost, and failure handling

Book Description:

Engineering a reliable AI system isn't as easy as calling an API. As AI applications move from prototypes to production, you must solve system-level problems the model alone cannot solve: retrieval quality, tool execution, hallucinations, validation, latency, cost, failures, observability, and operational control.

Systems Thinking for Agentic AI gives engineers and architects a practical framework for tackling these challenges. You'll first understand how LLMs work through tokens, embeddings, and transformers, then learn to control their behavior with prompting, structured outputs, and decoding strategies. You'll connect models to real systems using function calling, tools, and MCP, and build grounded RAG pipelines with embeddings and vector databases.

As you progress, you'll design agentic workflows with planning, memory, orchestration, and controlled execution. You'll implement guardrails, permissions, validation, human approval, evaluation, regression testing, and observability through logs, metrics, and traces. You'll also address production concerns including scaling, performance, cost, timeouts, retries, and fallbacks.

Finally, you'll bring these principles together by building an end-to-end Code Review Agent with Spring Boot, giving you a practical blueprint for engineering AI systems you can confidently operate in production.

What You Will Learn:

- Understand how tokens, embeddings, and transformers power LLMs

- Control LLM behavior with prompts and structured outputs

- Connect AI systems to tools using function calling and MCP

- Build grounded RAG pipelines with embeddings and vector databases

- Design agents using planning, memory, and orchestration

- Apply guardrails, validation, permissions, and human approval

- Evaluate and debug AI systems with tests, metrics, and traces

- Engineer for latency, cost, scaling, retries, and failures

Who this book is for:

This book is for technical leads, software architects, software engineers, and backend developers who want to move beyond AI prototypes and build reliable LLM and agentic AI systems for production. No machine learning background is required. Readers should be comfortable with backend development concepts such as APIs, distributed systems, and system design. Java and Spring Boot developers will also benefit from the end-to-end Code Review Agent implementation.

Table of Contents

- From AI to Agent Systems

- LLM Internals and Output Generation

- Prompt Engineering in LLM Systems

- LLMs as System Components

- RAG Systems (Retrieval Augmented Generation)

- Agent Fundamentals and Behavior

- Tool Orchestration and MCP

- Planning and Execution Control

- Memory in Agent Systems

- Agent System Architecture

- Reliability and Safety in AI Systems

- Performance and Cost Optimization

- Multi-Agent Systems

- Evaluation in Agent Systems

- Observability in AI Systems (Logs, Metrics, and Traces)

- End-to-End Agent System Design

- Implementing the Code Review Agent

- The Future of Agent Systems