GenAI on Google Cloud Enterprise Generative AI Systems and Agents (Ayo Adedeji, Lavi Nigam, Sarita Joshi etc.)（Z-Library）

Ayo Adedeji, Lavi Nigam, Sarita A. Joshi & Stephanie Gervasi GenAI on Google Cloud Enterprise Generative AI Systems and Agents

Praise for GenAI on Google Cloud We are moving past the era of “wow” moments in Generative AI into the era of “work.” This book provides the essential engineering blueprint for that transition. The authors don’t just talk about the potential of agents and LLMs; they show you exactly how to architect, secure, and scale them using the very best of Google Cloud’s stack. A mandatory read for any leader serious about turning AI experimentation into enterprise value. —Saurabh Tiwary, VP, general manager, Google Cloud AI A masterclass in balance, this book seamlessly bridges the gap between foundational theory and hands-on execution. For anyone looking to understand the “why” behind Generative AI while mastering the “how” of building applications, this is essential reading. —Burak Gokturk, VP of AI and systems research, Google Cloud Won’t any book about AI agents be woefully out of date as soon as it’s published? Perhaps for some, but not this one. The authors provide hearty treatment to a wide range of topics, focusing on the “why” as much as the “how.” Whether you’re creating AI strategy, designing architectures, writing code, or operating production systems, this book will stand the test of time as a valid guide towards a successful implementation. —Richard Seroter, chief evangelist, Google For leaders ready to build the next generation of intelligent applications, this book provides the necessary blueprint. It offers the first comprehensive look at Agent Development Kit (ADK), equipping engineers with the practical frameworks needed to turn agentic concepts into production-grade reality. —Julia Wiesinger, group product manager, Google

While ADK gives developers the framework to build powerful agents, this book provides the essential roadmap from building prototypes to secure, production-grade enterprise systems. It is the missing manual for anyone serious about bridging the gap between writing agentic code and delivering real-world business value on Google Cloud. —Bo Yang, lead software engineer, Google (Building ADK!)

Ayo Adedeji, Lavi Nigam, Sarita A. Joshi, and Stephanie Gervasi GenAI on Google Cloud Enterprise Generative AI Systems and Agents

979-8-341-62385-9 [LSI] GenAI on Google Cloud by Ayo Adedeji, Lavi Nigam, Sarita A. Joshi, and Stephanie Gervasi Copyright © 2026 Ayo Adedeji, Lavi Nigam, Sarita A. Joshi and Stephanie Gervasi. All rights reserved. Published by O’Reilly Media, Inc., 141 Stony Circle, Suite 195, Santa Rosa, CA 95401. O’Reilly books may be purchased for educational, business, or sales promotional use. Online editions are also available for most titles (http://oreilly.com). For more information, contact our corporate/institutional sales department: 800-998-9938 or corporate@oreilly.com. Acquisitions Editor: Nicole Butterfield Development Editor: Sara Hunter Production Editor: Aleeya Rahman Copyeditor: Stephanie English Proofreader: Dave Awl Indexer: Sue Klefstad Cover Designer: Susan Brown Cover Illustrator: José Marzan Jr. Interior Designer: David Futato Interior Illustrator: Kate Dullea January 2026: First Edition Revision History for the First Edition 2026-01-21: First Release See http://oreilly.com/catalog/errata.csp?isbn=9798341623859 for release details. The O’Reilly logo is a registered trademark of O’Reilly Media, Inc. GenAI on Google Cloud, the cover image, and related trade dress are trademarks of O’Reilly Media, Inc. The views expressed in this book are solely authors’ own and do not reflect the opinions of their employer or the publisher’s views. This content synthesizes public documentation for educational purposes only and contains no proprietary information or specific customer identifiers. While the publisher and the authors have used good faith efforts to ensure that the information and instructions contained in this work are accurate, the publisher and the authors disclaim all responsibility for errors or omissions, including without limitation responsibility for damages resulting from the use of or reliance on this work. Use of the information and instructions contained in this work is at your own risk. If any code samples or other technology this work contains or describes is subject to open source licenses or the intellectual property rights of others, it is your responsibility to ensure that your use thereof complies with such licenses and/or rights.

Table of Contents Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi 1. The Challenge of Generative AI Application Development. . . . . . . . . . . . . . . . . . . . . . . . . 1 Overview of LLMs, Generative AI Agents, and Potential Applications to Business Tasks 1 Small Language Models (SLMs) 4 Foundation Models and Multimodality 5 Domain-Specific and Reasoning Models 8 Generative AI Agents 10 Agent Architectures 13 Challenges in Development, Deployment, and Maintenance 15 Development Challenges 15 Deployment Challenges 18 Maintenance Challenges 20 Addressing Challenges with Modern Platforms 21 Industry Use Cases and ROI 22 Looking Ahead 24 Learning Labs 25 2. Data Readiness and Accessibility. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 The Amplified Importance of Data for GenAI 27 What Data Readiness Really Means for GenAI Applications 29 Key Dimensions of Data Readiness 30 The Interconnected Nature of Data Readiness 33 Managing Prompts as Data Assets 34 The Human Element: Roles in the Data Readiness Journey 35 Data Scientists: The Explorers 35 ML Engineers: Building the Bridge to Production 37 v

Data Engineers: Architecting the Foundation 38 DevOps and SREs: Operationalizing the Foundation 39 Business SMEs and Domain Leaders: The “Why” Behind the “What” 40 Strategic Data Patterns: The Foundation for Reliable GenAI Systems 41 The Unified Data and AI Platform 41 From RAG to Agentic RAG: The Evolution of a Data Pattern 45 Tying it All Together: the Enterprise RAG Knowledge Engine 55 Data Readiness for Agent Systems 58 Security and Governance: Protecting Data Throughout the LLM Lifecycle 60 Data Privacy Framework 60 Comprehensive Governance 62 Practical Data Readiness Assessment 63 Looking Ahead 66 Learning Labs 67 3. Building a Multimodal Agent with the Agent Development Kit (ADK). . . . . . . . . . . . . . 69 From Zero to Agent in Seven Lines 71 The Simplest Thing That Works 71 The Runtime Behind the Simplicity 72 Running Your First Conversation 73 Understanding the Limitations 74 Adding Intelligence Through Tools 74 Your Agent’s First Tool 75 Tools Versus Subagents–A Practical Decision Framework 79 State Management That Actually Scales 82 Building a Stateful Shopping Cart 82 Understanding the Three Scopes 85 State Scope Interactions 86 Making State Persist in Production 87 Beyond Structured State: Semantic Memory 89 Vertex AI Agent Engine Memory Bank: Learning from Conversations 90 Implementation 91 Expanding to Multimodal 93 Making Our Agent See 94 From Static Analysis to Live Support 96 Building Complete Interaction Memory 100 Building Production-Grade Tools 101 Handling Asynchronous Operations 101 Ensuring Safety with Human-in-the-Loop 103 Production Monitoring and Policy Enforcement with Callbacks and Plug-ins 105 Looking Ahead 108 Learning Labs 109 vi | Table of Contents

4. Orchestrating Intelligent Agent Teams. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111 The Bottleneck of the Monolithic Agent 111 Conflicting Instructions 112 Tool Selection Paralysis 112 Token Limitations 113 Maintenance Nightmare 113 The Solution: An Agent Team 114 The Roadmap: From Local Teams to Distributed Systems 115 Local Teams 116 The Foundation: Agent Hierarchy 116 Pattern 1: The Assembly Line (SequentialAgent) 117 Pattern 2: The Independent Taskforce (ParallelAgent) 119 Pattern 3: The Iterative Refiner (LoopAgent) 121 Distributed Collaboration 124 The Organizational “Why” 124 MCP: The Language of Tools 127 A2A: The Language of Delegation 130 Putting It All Together: A Hybrid Agent Team 134 Production Realities 137 The Trust Problem: Security Schemes in A2A 138 The Extension Problem: Evolving Agent Capabilities 139 The Visibility Problem: Distributed Tracing 141 The Versioning Problem: Managing Agent Evolution 142 Looking Ahead 143 Edge and Embodied Intelligence 143 From Architecture to Excellence 143 Learning Labs 144 5. Evaluation and Optimization Strategies. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 Tailoring Evaluation to Your LLM/Agent’s Purpose 148 Beyond Basic Functionality 148 Key Dimensions of Evaluation 149 Setting the Bar for Production Excellence 153 Practical Evaluation Strategies 154 Human-Centered Evaluation 154 A/B Testing and Preference Scoring 156 Red Teaming: Stress Testing for Safety and Reliability 159 Automated Evaluation: Scaling Feedback for Rapid Improvement 160 Reference-Based Metrics for Text Generation 160 Limitations of Reference-Based Evaluation 161 Domain-Specific and Task-Oriented Metrics 162 Metrics for Agentic Systems and Tool Use 162 Table of Contents | vii

Optimization Strategies 169 Refining Prompts 169 Elevating Agent Performance 171 Beyond Prompt and Agent Optimizations 173 Looking Ahead 174 Learning Labs 175 6. Tuning and Infrastructure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177 The Tuning Decision 177 The Fine-Tuning Decision Framework 178 Fine-Tuning Strategies: From Full Training to Efficient Adaptations 180 The Real Cost of Fine-Tuning 181 Implementation Approaches 181 Infrastructure Questions Emerge 184 The Constraint You’ll Hit First 184 Pattern 1: The Waiting Accelerator 185 Pattern 2: The Memory Wall 186 Pattern 3: Maxed Out But Still Slow 187 Pattern 4: More GPUs = Worse Performance 188 Accelerators: Matching Hardware to Bottlenecks 189 The Decision Framework 189 The Practical Decision 190 Migration Reality 191 Storage Options 191 When Storage Becomes Your Bottleneck 192 The Storage Pattern 192 Serving and Deployment 193 Configuration That Matters 195 Connecting Models to Agents 196 Agent Deployment Platforms 197 Agent Engine 198 Cloud Run 199 GKE 200 Looking Ahead 202 Learning Labs 203 7. MLOps for Production-Ready AI and Agentic Systems. . . . . . . . . . . . . . . . . . . . . . . . . . . 205 From Ad Hoc to Systematic: The Current State of Teams 207 The Evolution of MLOps 208 Building Reproducible Training Pipelines 209 Data Versioning and Lineage 209 Experiment Tracking 213 Model Registry and Governance 214 viii | Table of Contents

Automated Retraining 217 Comprehensive Monitoring 219 Agent Monitoring 219 Technical Monitoring 222 Hallucination Detection 224 CI/CD for AI Systems 225 Cloud Build 225 Cloud Deploy 226 Security and Governance as Foundation 227 Security Framework for AI Agents 227 Model Armor: A Key Security Component 227 Cost Management 228 The True Cost Model 229 Cost Attribution Strategies 230 Intelligent Cost Operations 231 Spending Controls 233 Looking Ahead 235 Learning Labs 236 8. The AI and Agentic Maturity Framework. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237 What Is the AI and Agentic Maturity Framework? 238 The Maturity Dimensions and Phases 238 Vision and Leadership (The “What” and the “Why” Dimension) 240 Talent and Culture (The “Who” Dimension) 246 Operational and Technical Practice (The “How” Dimension) 250 How the Three Dimensions of AI and Agentic Maturity Can Work Together 261 From Framework to Reality: What Are Teams Actually Building, and How? 263 Technical Conversations 263 Leadership, Talent, and Culture Conversations 264 Why and How a Platform Approach Can Accelerate an Organization’s AI and Agentic Maturity 265 Vertex AI Platform 266 Learning Labs 269 Conclusion. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271 Appendix. Further Reading for Leaders. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 275 Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277 Table of Contents | ix

Preface Let’s be honest—building a flashy generative AI prototype is the easy part. We’ve all been there: the excitement of a demo that works perfectly in controlled conditions, stakeholders leaning forward in their chairs, and that moment when someone says, “This is amazing! When can we roll it out?” And then reality hits. The four of us—Ayo, Lavi, Sarita, and Steph—have collectively spent thousands of hours helping organizations and developers navigate what comes after that excit‐ ing demo moment. We work across different teams at Google Cloud, but we kept encountering the same fundamental challenge regardless of industry or use case: the gap between a working prototype and a production-ready generative AI system is massive, complex, and filled with obstacles that aren’t obvious until you’re knee-deep in the journey. During one of our regular knowledge-sharing sessions, we realized that while there are plenty of resources that teach how to build generative AI prototypes, there’s sur‐ prisingly little practical guidance on the critical path from prototype to production. We started compiling our notes, frameworks, and hard-earned lessons, initially just for ourselves and our teams. “This could help a lot more people than just us,” became our mantra, and that’s how this book was born. We’re not here to dazzle you with theoretical abstractions or rehash concepts you can find in a hundred blog posts. Instead, we’ve created the practical guide we wish we’d had—one that addresses the real challenges of deploying generative AI systems in production environments on Google Cloud, backed by concrete examples and honest insights about what actually works. This isn’t a polished marketing narrative. It’s a field guide written by practitioners who’ve seen both the triumphs and the train wrecks. We’ve kept our individual voices throughout, sharing our unique perspectives rather than forcing a unified narrative, xi

because that’s how real engineering teams work. Sometimes we disagree, and that’s valuable too. Our goal is simple: to help you navigate the journey from prototype to production more efficiently than we did our first time around. Why This Book Matters The GenAI and agentic AI landscape is evolving at breakneck speed. New models emerge almost daily, each promising more parameters, better benchmarks, and greater capabilities. Yet across industries, we’re witnessing a consistent pattern: organ‐ izations struggle not with building impressive prototypes, but with making them production-ready. This transition from prototype to production represents the true value inflection point. According to industry analysts, more than 80% of enterprises will have experi‐ mented with generative AI by 2026, yet many report no significant bottom-line impact. At the heart of this paradox is a mismatch in deployment. While “horizontal” tools like chatbots have scaled rapidly across enterprises, their value is often spread thin and difficult to quantify. In contrast, high-value “vertical” applications—those deeply integrated into specific business functions—face far greater hurdles, with analysts estimating that nearly 90% of these transformative initiatives fail to progress beyond the pilot stage. The gap is particularly stark in regulated industries such as healthcare and financial services, where security, compliance, and reliability require‐ ments intensify the challenge. What separates successful deployments from a pilot experiment? It’s rarely about model quality alone. Instead, it’s about the surrounding infrastructure—data pipe‐ lines, evaluation frameworks, monitoring systems, and governance guardrails—that make the difference between systems that demonstrate potential and those that deliver measurable business value. Google Cloud’s Vertex AI platform has helped hundreds of organizations bridge this gap, providing the foundation for both startups and Fortune 500 companies to move their GenAI applications from concept to production. The patterns and practices we’ve observed along the way form the backbone of this book’s practical approach to building systems that not only work in demos but thrive in the real world. Writing a book about a field that reinvents itself every few months is a formidable challenge. While specific code snippets, tools, or model recommendations are bound to become dated, the fundamental questions of system design, evaluation, and gover‐ nance will remain. Our goal is to equip you with a durable framework for answering those questions, regardless of what the next breakthrough looks like. xii | Preface

What You’ll Find in This Book We’ve structured this book to follow the journey from prototype to production, with each chapter building on the previous one: Chapter 1, “The Challenge of Generative AI Application Development” Introduces the unique complexities of building GenAI applications and sets the stage for MLOps for LLMs and agents Chapter 2, “Data Readiness and Accessibility” Explores how to handle and prepare data for GenAI applications, covering prep‐ aration strategies, annotation techniques, and data augmentation Chapter 3, “Building a Multimodal Agent with the Agent Development Kit (ADK)” Guides you through creating a functional agent prototype using Google’s ADK framework, demonstrating how to build a customer service agent that handles text, video, and images Chapter 4, “Orchestrating Intelligent Agent Teams” Shows how to scale from single agents to multiagent systems, covering agent-to- agent and agent-to-tool communication patterns and enterprise collaboration strategies Chapter 5, “Evaluation and Optimization Strategies” Explores frameworks for measuring success in LLMs and agents, from safety and robustness metrics to agent trajectory evaluation and advanced optimization patterns Chapter 6, “Tuning and Infrastructure” Covers model fine-tuning techniques and infrastructure optimization strategies to maximize performance and cost-efficiency in production environments Chapter 7, “MLOps for Production-Ready AI and Agentic Systems” Establishes a comprehensive approach to operationalizing LLMs and agent sys‐ tems, from CI/CD pipelines and cost management to monitoring and deploy‐ ment strategies Chapter 8, “The AI and Agentic Maturity Framework” Provides a strategic roadmap for organizational growth, helping you assess readi‐ ness across leadership, culture, and operations to move from tactical experiments to transformational AI adoption Each chapter concludes with a “Looking Ahead” section to prepare you for the topics and themes that will come next. Additionally, we provide a “Learning Labs” section at the end of each chapter, where links to hands-on activities will lead you to further support your understanding of key concepts presented in the chapter. Preface | xiii

Our Approach We believe in learning by doing. Throughout this book, we provide code examples that you can run and adapt to your specific needs. We focus on practical implementa‐ tions rather than theoretical abstractions, though we provide enough theory to ensure that you understand why certain approaches work better than others. We’ve chosen to write this book with our individual voices rather than aiming for a seamless narrative. As you read, you’ll hear from each of us directly, sharing our specific expertise and experiences. We believe this approach makes the content more authentic and allows us to connect with you on a more personal level. Who This Book Is For This book is designed for several key audiences: • Machine learning engineers and AI engineers transitioning from traditional machine learning models to complex generative AI pipelines • Data teams moving from conventional analytics to AI-powered insights • Software developers with Python skills entering AI-first application development • Product managers and technical leaders responsible for AI strategy and implementation • Career transitioners leveraging existing technical foundations to move into AI engineering roles While we assume familiarity with Python programming and basic machine learning concepts, we’ve structured the content to be accessible to readers with varying levels of expertise. Some familiarity with Google Cloud and Vertex AI is beneficial but not a prerequisite. Prerequisites To get the most out of this book, you should have: • Experience with Python programming • Basic understanding of machine learning concepts • Familiarity with cloud computing principles (though not necessarily Google Cloud specifically) If you’re new to some of these areas, don’t worry—we provide references and explan‐ ations where needed, and the hands-on approach means you’ll learn as you go. xiv | Preface

Conventions Used in This Book The following typographical conventions are used in this book: Italic Indicates new terms, URLs, email addresses, filenames, and file extensions. Constant width Used for program listings, as well as within paragraphs to refer to program elements such as variable or function names, databases, data types, environment variables, statements, and keywords. Constant width bold Shows commands or other text that should be typed literally by the user. Constant width italic Shows text that should be replaced with user-supplied values or by values deter‐ mined by context. This element signifies a general note. Using Code Examples Supplemental material (code examples, exercises, etc.) is available for download at https://github.com/ayoisio/genai-on-google-cloud. If you have a technical question or a problem using the code examples, please send email to support@oreilly.com. This book is here to help you get your job done. In general, if example code is offered with this book, you may use it in your programs and documentation. You do not need to contact us for permission unless you’re reproducing a significant portion of the code. For example, writing a program that uses several chunks of code from this book does not require permission. Selling or distributing examples from O’Reilly books does require permission. Answering a question by citing this book and quoting example code does not require permission. Incorporating a significant amount of example code from this book into your product’s documentation does require permission. We appreciate, but generally do not require, attribution. An attribution usually includes the title, author, publisher, and ISBN. For example: “GenAI on Google Cloud: Enterprise Generative AI Systems and Agents by Ayo Adedeji, Lavi Nigam, Sarita A. Preface | xv

Joshi, and Stephanie Gervasi (O’Reilly). Copyright 2026 Ayo Adedeji, Lavi Nigam, Sarita A. Joshi, and Stephanie Gervasi, 979-8-341-62385-9.” If you feel your use of code examples falls outside fair use or the permission given above, feel free to contact us at permissions@oreilly.com. O’Reilly Online Learning For more than 40 years, O’Reilly Media has provided technol‐ ogy and business training, knowledge, and insight to help companies succeed. Our unique network of experts and innovators share their knowledge and expertise through books, articles, and our online learning platform. O’Reilly’s online learning platform gives you on-demand access to live training courses, in-depth learning paths, interactive coding environments, and a vast collection of text and video from O’Reilly and 200+ other publishers. For more information, visit https://oreilly.com. How to Contact Us Please address comments and questions concerning this book to the publisher: O’Reilly Media, Inc. 141 Stony Circle, Suite 195 Santa Rosa, CA 95401 800-889-8969 (in the United States or Canada) 707-827-7019 (international or local) 707-829-0104 (fax) support@oreilly.com https://oreilly.com/about/contact.html We have a web page for this book, where we list errata, examples, and any additional information. You can access this page at https://oreil.ly/GenAI_on_Google. For news and information about our books and courses, visit https://oreilly.com. Find us on LinkedIn: https://linkedin.com/company/oreilly-media. Watch us on YouTube: https://youtube.com/oreillymedia. xvi | Preface

Acknowledgments Ayo I’d like to thank my dad, mom, sister, and friends for their patience and engage‐ ment as I’ve subjected them to countless philosophical inquiries about AI and the future of, well, everything. Through our conversations—sometimes skeptical, always thoughtful—they’ve taught me how different people see the world and what they value. This understanding has been invaluable in writing a book meant to serve diverse readers with varied perspectives and concerns about AI. To O’Reilly Media, thank you for the honor and trust of this platform. The opportunity to teach and inspire curiosity about AI—to help others experience that same wonder I felt when I first saw an LLM understand context or an agent solve a complex problem—is truly a gift. Lavi This book was written for the builder. My heartfelt thanks to everyone who helped bring it to life. To my family, my foundation and constant source of strength. To my mom and dad, thank you for the incredible privilege of a life shaped by your unwavering support, a solid education, and the belief that I could achieve anything. Your sup‐ port has been my greatest advantage. To my sister, brother, and all siblings–your constant encouragement has been a true gift. To my entire family, thank you for always making me believe that I can solve anything and do great things. My deepest gratitude extends to my friends and all the wonderful colleagues I have had and currently have at Google, who have always bestowed their trust in me and cheered me on through every chapter of my life. I am also profoundly grateful for the journey that led me here. My school and college education laid the groundwork for the person I am today, and I am indebted to all my previous and current managers and my leadership, who have mentored me, challenged me, and helped build me into the professional I have become. To my colleagues at Google, thank you for building the tools and infrastructure that are changing our industry and for being an incredible source of inspiration. I am also humbled by the broader AI community; your passion fuels my own. Your relentless innovation is what makes this guide relevant. I am also pro‐ foundly grateful to the global developer community. Your curiosity, challenging questions, and passion for building are the driving force behind my work. To my coauthors, Ayo, Sarita, and Steph: I am incredibly proud of what we built. This book is a testament to our shared commitment to helping others succeed in moving from prototype to production. Thank you for your partnership. Preface | xvii

This book is a milestone in a mission I care about deeply: making complex AI accessible to every developer and enterprise. This work is impossible alone. Sarita Writing this book has been a profound reminder that no achievement is solitary. My deepest gratitude begins with my family and friends, whose belief in me has been a constant anchor, no matter the physical distance. A special acknowledge‐ ment goes to my elder brother, Kishor Joshi, who many years ago used his first paycheck to buy me my first personal laptop. That single gift opened the door to the world of technology for me long before a book like this was even imaginable, and I am forever grateful. Professionally, I am driven by the people I serve and work with. To my customers and the visionary leaders in healthtech: thank you for inspiring me daily. Your complex challenges are the motivation behind the practical solutions in this book. Your mission is my mission. And to my colleagues at Google—the countless innovators and experts across multiple teams who I am lucky to call friends: your intellectual curiosity and relentless support made this project possible. This work is a testament to our collaboration. Stephanie This book is the result of the cumulative wisdom, patience, and support drawn from many people. To every individual who contributed to my journey, directly or indirectly, I offer my heartfelt gratitude. I am grateful for the foundational community of my family and friends. Your belief in me, your encouragement, and your understanding made it possible for me to undertake this project. I extend my gratitude to my amazing Google collea‐ gues, professional peers, and all of the students, researchers, and technologists I’ve had the pleasure of meeting and talking with over these last years. The idea and execution of this book was born out of intellectual curiosity, collaboration, and genuine excitement for the field of AI, with all the potential it holds to improve people’s lives. I dedicate a specific and vital thank-you to the customers and stakeholders I have had the privilege of working with. This book exists because of the real-world complexities you bring to the table. Your trust in my counsel, and the honest details of your operational challenges have served as the ultimate proving ground for these ideas. A final dedicated space must be reserved for the incredible collaboration that defined this book: my coauthors. More than just partners in writing, you were the essential sounding boards, technical anchors, and creative force that ensured this vision was realized. Thank you for your tireless effort, intellectual rigor, and shared commitment to seeing this project through every demanding stage. It has been such a joy to work with you on this book! xviii | Preface

Ayo, Lavi, Sarita, and Steph would like to specially thank our development editor, Sara Hunter, for her guidance and support throughout this project, as well as the entire team at O’Reilly who helped bring this book to life. We’re also grateful to Google Cloud for providing the platform and opportunity to work with so many organizations on their GenAI journeys, which provided the insights and experiences that form the foundation of this book. A big thanks to subject mat‐ ter experts who reviewed drafts of our book, including Saurabh Tiwary, Burak Gokturk, Ashok Rao, Yasmeen Ahmad, Jason Gelman, Alan Blount, Bo Yang, Julia Wiesinger, Christina Lin, Polong Lin, Dave Elliott, Tanya Singh, Irina Sigler, Blane Clark, and Dr. Pete Clardy. Most importantly, we want to collectively thank the developers, engineers, and organizations who have shared their challenges and successes with us. Your ques‐ tions, feedback, and innovative solutions have not only made this book possible but continue to drive the evolution of GenAI on Google Cloud. This book is for you. Preface | xix