Librería Samer Atenea
Kálamo Books
Librería Elías (Asturias)
Librería Kolima (Madrid)
Librería Proteo (Málaga)
Agentic AI Systems: Architectures and Design Patterns for Autonomous IntelligenceMost developers who work with large language models hit the same wall eventually. The model is impressive. The demos work. Then something breaks in production at two in the morning and there is no architectural foundation to fall back on, because what was built was never really a system. It was a prompt chain dressed up to look like one.This book is about closing that gap. Not the gap between knowing about AI and understanding AI, but the more practical gap between having access to a capable language model and having built something that runs autonomously, handles failures gracefully, and does useful work without someone driving every step.The difference between a language model and an agent sounds simple on paper. A language model takes input, produces output, and stops. An agent has a goal. It takes steps toward that goal, uses tools to affect the world around it, monitors what happens, and adjusts when things go wrong. That architectural difference turns out to change almost everything about how you design, build, test, and operate the system.This is a practitioner’s book written by someone who has written thousands of lines of LangGraph orchestration code, failed at AutoGen fine-tuning more than once, and stood in front of product teams trying to explain what 'reflection' means in the context of an AI agent. The frameworks covered here - LangGraph, CrewAI, AutoGen - exist today, ship to production today, and are being used by real companies to automate real work. The research literature is valuable, but this book lives in the space between theory and deployment.The book is organized in four sections. The first covers the foundations: what an agent actually is, how memory and planning work, where the major generative model architectures fit in, and why the transformer became the engine underneath all of it. The second gets into architecture - reactive agents, deliberative agents, hybrid systems, the Coordinator-Worker-Delegator pattern, tool use, and the underappreciated complexity of state management. The third covers scale and operations: multi-agent coordination, evaluation, human-in-the-loop design, and governance. The fourth closes with applied case studies and design canvases drawn from production deployments in marketing, financial research, healthcare navigation, and DevOps automation.Throughout, the code examples are concrete and runnable, not illustrative pseudocode. The architectural principles are grounded in what actually breaks in production, not in what looks elegant on a whiteboard. And the honest parts - the places where the tooling is still immature, where the models are unreliable in specific ways, where the governance questions do not have clean answers yet - are included because pretending otherwise does not help anyone build better systems.If you are a developer who has gotten reasonably good at prompt engineering and wants to understand what comes next, this book is for you. If you are a technical lead trying to figure out what 'adding AI' to a product actually requires architecturally, this book is for you. If you are building something autonomous and want a clear-eyed guide to the patterns that work and the failure modes you will encounter, this is the book.The field is moving fast. Some of what is written here will be outdated by the time you read it. What will not be outdated is why you need memory, why planning matters, why human oversight is not optional, and why an agent that cannot handle failure gracefully is not really an agent at all. Those principles hold regardless of which framework is newest. Everything else you can adapt.
