Agentic Mesh The GenAI-Powered Autonomous Agent Ecosystem (Eric Broda, Davis Broda) (z-library.sk, 1lib.sk, z-lib.sk)

Agentic Mesh The GenAI-Powered Autonomous Agent Ecosystem Eric Broda & Davis Broda Foreword by Sean Falconer

ISBN: 979-8-341-62164-0 US $79.99 CAN $99.99 DATA With the rise of autonomous agents, the question is no longer “How do we build agents?” but rather, “How do we manage an entire ecosystem of them?” This book explores the next frontier of this technology, where interconnected agents collaborate, transact, and fulfill tasks autonomously. Building on established concepts like API service mesh and data mesh, authors Eric and Davis Broda introduce agentic mesh as a transformative architecture designed to safely manage growing ecosystems of agents at scale. This practical guide unpacks how agentic mesh works, explains its key components—such as agent registries, marketplaces, trust-building mechanisms, and human-in-the-loop oversight—and illustrates how agents can discover, interact, and transact with ease. Through accessible explanations and compelling use cases, you’ll gain a clear understanding of how to implement and govern agent ecosystems, ensuring security, transparency, and efficiency. Whether you’re a tech leader, developer, or enterprise strategist, Agentic Mesh provides a road map to navigate the future of autonomous agents with confidence. • Understand the concept of agentic mesh and its transformative potential • Learn how autonomous agents find, collaborate, and transact within a mesh ecosystem • Explore critical components like agent registries, marketplaces, and trust frameworks • Address safety, security, and governance challenges in agent ecosystems • Apply practical use cases to begin designing and implementing your own agentic mesh Eric Broda is a technology executive, author, and founder of a consulting firm that helps global enterprises accelerate their agent journey. He writes on the emerging agent ecosystem, and cowrote O’Reilly’s Implementing Data Mesh. Davis Broda is a senior software architect and engineer with experience leading technology teams at banks and tech firms. He develops generative AI, security, and application solutions, and works with tools like CrewAI, LangGraph, and OpenAI Swarm. Agentic Mesh “Agentic Mesh provides the enterprise architecture blueprint we’ve been missing. This is a definitive guide for architects and executives navigating the shift from isolated AI LLM and agent experiments to coordinated agent ecosystems.” Kerrie Holley, IBM fellow, member of the National Academy of Engineering, and National Inventors Hall of Fame inductee “We are moving rapidly from siloed AI to sophisticated agentic architectures. This book is a masterclass for how companies can build a connective tissue for AI. Essential reading!” Bruno Aziza, GVP of enterprise software, IBM

Praise for Agentic Mesh The future of AI will be shaped less by individual models and more by the systems that connect them. Agentic Mesh is one of the clearest and most thoughtful treatments of what those systems must look like, offering a rigorous, practical framework for building agent ecosystems that can operate reliably at enterprise scale. —Sean Falconer, head of AI, Confluent Agentic architectures are on the rise; as these architectures mature, we will see them emerge as a mesh of agents working together. This book provides highly needed guidance for such a reality. —Ole Olesen-Bagneux, chief evangelist, Actian, and O’Reilly author of Fundamentals of Metadata Management and The Enterprise Data Catalog Agentic Mesh provides a badly needed introduction to autonomous AI agents. The book starts out by explaining the basic concepts of AI agents and provides a practical guide to how to move from simple, autonomous agents toward increasingly large fleets of agents and enterprise-grade agentic mesh ecosystems. —Irving Wladawsky-Berger, research affiliate, MIT, and former senior executive, IBM In the time of perpetual movement and acceleration of tech, a book may appear and become immediately obsolete. This is definitely not the case for this book. Agentic Mesh not only sets an enterprise-grade foundation for agentic AI but also draws a picture of the future. Highly recommended. —Jean-Georges Perrin, data and AI strategist, Actian

Agentic Mesh provides the enterprise architecture blueprint we’ve been missing. Eric and Davis Broda cut through the AI hype to deliver a practical, production-ready framework for scaling autonomous agents across the organization. This is a definitive guide for architects and executives navigating the shift from isolated AI LLM and agent experiments to coordinated agent ecosystems. —Kerrie Holley, IBM fellow, member of the National Academy of Engineering, and National Inventors Hall of Fame inductee Agentic AI is not like previous technologies—it’s about creating near-infinite digital workforces to complement human workers, at near zero marginal cost. As such, the stakes are unprecedented and enterprise architecture—the connection between strategy and technology—is so important. Agentic Mesh is a critical read for all those looking to win with Agentic AI. —Simon Torrance, CEO, AI Risk Agentic Mesh explains agentic AI through a chief architect’s lens—layer by layer, with the right abstractions in the right places. It offers a practical mental model for building reliable, governed systems that can be trusted as agentic AI scales from experimentation to enterprise reality. —John Y. Miller, data and AI strategist, former Accenture chief data architect and R&D lead We are moving rapidly from siloed AI to sophisticated agentic architectures. To move from theory to what I call systems of action, we need a rigorous framework for how these agents interact, govern, and collaborate. Agentic Mesh is a masterclass for how companies can build a connective tissue for AI. Essential reading! —Bruno Aziza, GVP enterprise software, IBM

Eric Broda and Davis Broda Foreword by Sean Falconer Agentic Mesh The GenAI-Powered Autonomous Agent Ecosystem

979-8-341-62164-0 [LSI] Agentic Mesh by Eric Broda and Davis Broda Copyright © 2026 Broda Group Software Inc. and Davis Broda. 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: Aaron Black Development Editor: Corbin Collins Production Editor: Elizabeth Faerm Copyeditor: nSight, Inc. Proofreader: Tim Stewart Indexer: Judith McConville Cover Designer: Susan Brown Cover Illustrator: Susan Thompson Interior Designer: David Futato Interior Illustrator: Kate Dullea February 2026: First Edition Revision History for the First Edition 2026-02-04: First Release See http://oreilly.com/catalog/errata.csp?isbn=9798341621640 for release details. The O’Reilly logo is a registered trademark of O’Reilly Media, Inc. Agentic Mesh, the cover image, and related trade dress are trademarks of O’Reilly Media, Inc. The views expressed in this work are those of the authors and do not represent the publisher’s views. 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 Foreword. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv Preface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xix Part I. Defining the Essentials 1. Understanding Agentic Mesh: The Essentials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 The Introduction of LLMs 4 The Agent Era 6 Defining Agents 8 Agents Today 9 Enterprise-Grade Agents 11 Agentic Mesh: The Agent Ecosystem 12 The Agent Challenge 15 The Agent Opportunity 16 Summary 17 2. Agentic Past, Present, and Future. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 The Agent Past 20 The Origins of Artificial Intelligence 20 The Era of Machine Learning 22 The Deep Learning Revolution 22 The Agent Present 23 The Transformer Architecture 23 The Age of LLMs 24 The Agentic Future 26 Short Term: Laying the Foundation for Enterprise-Grade Agents 26 v

Medium Term: The Rise of Agentic Mesh—the Agent Ecosystem 28 Long Term: The Creation of the Agent Businesses 29 Summary 31 3. Agents Versus AI Workflow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 Defining AI Workflows 33 Common Types of AI Workflows 34 Prompt Chaining 34 Routing 37 Parallelization 38 Orchestration 39 Reflection 40 Challenges with AI Workflows 42 The “Black Box” Issue 42 Scaling Challenges 43 Handling Edge Cases 44 Comparing AI Workflows and Agents 45 Agents Extend AI Workflows 46 Summary 48 4. Agent Basics. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 Agent Analogy: Agents as People 50 From Person to Agent 50 From Teams to Agent Fleets 52 From Organizations to Agent Ecosystems 53 Architecture of an Agent 55 Task Planning 57 Task Execution 59 Problem-Solving 62 Tool Use 63 Memory and Context 65 Learning 67 Collaboration and Communications 69 Summary 71 Part II. Defining the Agent Ecosystem: Agentic Mesh 5. Agent Architecture. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 Agent Principles 76 Trustworthy and Accountable 78 Reliable and Durable 80 vi | Table of Contents

Explainable and Traceable 81 Collaborative and Intelligent 84 Agent Components 86 Agent “Brain” 87 Agent Memory 88 Agent Context Engineering 89 Agent Tools 92 Agent Task Management 93 Creating the Task Plan 94 Identifying Collaborators and Tools 96 Parameters Substitution 97 Executing the Task Plan 98 Agent Interactions and Conversations 99 Agent Messaging Model 99 Agent Conversation Management 102 Agent State Management 103 Agent Workspaces 105 Agent Identities and Roles 106 Agent Types 107 Task-Oriented Agents 107 Goal-Oriented Agents 108 Simulation Agents 110 Observer Agents 112 Agent Patterns 113 Agent Communication Patterns 113 Agent Role Patterns 116 Agent Organizational Patterns 118 Agent Configuration 121 Identity, Description, and Purpose 123 Task Execution Strategy 124 Security Configuration 124 Policy and Certification 125 Agent and Tool Visibility 125 Summary 126 6. Enterprise-Grade Agents. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127 Microagents (Microservice Agents) 128 Agent Security 130 Basic Microservices Security 131 Container Security 132 Kubernetes Security 133 Agent Reliability 134 Table of Contents | vii

The Reliability Problem 134 The Reliability Problem Root Cause 135 Potential Solutions 136 Task Decomposition 139 Deterministic Execution 139 Specialization 141 A Future with Reliable Agents 142 Agent Explainability 143 The Trust Gap 143 Explainability: Real-World Lessons 144 Explaining Explainability 145 Toward Explainable Agents 148 Agent Scalability 148 The Scalability Problem 149 Distributed Architectures 152 Common Collaboration Techniques 153 Conversation/State Management 155 Enterprise-Grade Agent Capabilities 157 Agents as the Quantum of Reuse 158 Scaling the LLM Foundation 160 Agent Discovery 161 Beyond a Search Problem 161 Finding the Right Agent 162 Agent Observability and Traceability 164 Agent Operability 165 Agent Testing 166 Testing LLMs 167 Extending to Agent Testing 167 Testing Microagents: Determinism Within the Probabilistic 168 AgentOps: DevOps for Agents 168 Summary 170 7. Agentic Mesh: Enterprise-Grade Agent Ecosystem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171 Ecosystems and Scale 172 Agent Fleets 173 Structure and Composition 173 Coordination and Operation 174 The Ecosystem Management Plane 176 Agentic Mesh Components 177 The Registry 178 The Monitor 181 The Interactions Server 182 viii | Table of Contents

The Marketplace 183 Workbenches 185 The Proxy 185 Mesh Capabilities 186 Trust Framework 186 Operations 187 Agent Lifecycle Management 188 Summary 189 8. Agentic Mesh User Experience (UX). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191 Agentic Mesh UX 193 Login 194 Home 195 Marketplace 196 What Is an Agent Marketplace? 197 Marketplace Services 198 Finding the Right Agent 202 Natural Language Search 202 Hierarchical Agent Navigation 202 Consumer Workbench: Engaging an Agent 203 Shared Workspaces for Agents 203 Chat Interfaces for Agents 206 Creator Workbench 209 Registering Agent Metadata 209 Connecting Agents to the Marketplace 210 Using Workbenches to Manage Agent Lifecycle 210 Similarities Between Creator Workbench and PyPI 211 Trust Workbench 211 Policy Configuration 212 Policy Attachment to Agents 212 Agent Certification 213 Internal and Third-Party Certification 213 Certification Lifecycle Management 214 Operator Workbench 214 Agent Observability 215 Diagnostics and Troubleshooting 215 Execution Control 215 Summary 216 9. Agentic Mesh Registry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217 Agentic Mesh Registry 218 Agents 219 Table of Contents | ix

Conversations 222 Interactions 224 Workspaces 226 Policies 228 Users 230 Implementation Considerations 232 Summary 233 10. Interaction Management. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235 Agentic Mesh Interaction Management 236 Event-Driven Communication 237 HTTP Versus Event-Driven 237 Message Queues: Reliable Delivery and Persistence 238 Pub/Sub: Dynamic and Scalable Distribution 239 Event Replay 240 Monitoring Queues 241 User-to-Agent Communication 242 Interaction Lifecycle 244 Conversations 246 Conversation and Interaction Endpoints 247 Agent-to-Agent Communication 247 Agents as Plan Steps 247 Sending Messages 248 User Alerts 249 Workspaces 251 Deciding to Respond 252 Acting on Workspace Messages 253 Workspace Goals 254 Workspaces as a Super-Context 256 Summary 257 11. Security Considerations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259 Agentic Mesh Security 260 Mutual TLS 261 Authentication and Authorization 262 Secrets Management 265 Prompt Injection 267 Prompt Injection Example 270 Techniques of Prompt Injection 271 Combating Prompt Injection 273 LLM Jailbreaking 274 Behavioral Monitoring 276 x | Table of Contents

Disaster Recovery 276 Summary 277 12. Trust Framework and Governance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279 Seven-Layer Agent Trust Framework 280 Layer 1: Identity and Authentication 283 Managing Identity Lifecycle 283 Delegating and Scoping Authority 283 Scaling Identity 284 Monitoring and Auditing Identity 284 Trust and Identities 284 Layer 2: Authorization and Access Control 284 Access Control Foundations 285 A Zero-Trust Model for Agents 285 Enforcement, Least Privilege, and Lifecycle Management 285 Identity Integration and Federated Governance 286 Operationalizing Security by Design 286 Layer 3: Purpose and Policies 287 Purpose: Defining What an Agent Does 287 Policies: Defining Operational Constraints 288 Layer 4: Task Planning and Explainability 289 The Problem: Opaque Reasoning in Today’s Agents 289 Choosing Tools and Collaborators 290 Parameterization and Step Execution 290 Layer 5: Observability and Traceability 290 Visibility into Agent Activity 291 Capturing Multiagent Task Contexts 291 Operational Monitoring and System-Level Observability 292 Layer 6: Certification and Compliance 292 Certification as Structured Assurance 293 Evaluation, Oversight, and Recertification 293 Trust Registries, Metadata, and Discoverability 294 Feedback Loops, Enforcement, and Long-Term Trust 295 Layer 7: Governance and Lifecycle Management 295 Agent Governance in Practice 295 Agent Lifecycle Management Implications 297 Summary 299 Table of Contents | xi

Part III. Building Your Agentic Mesh 13. Operating Model and Team Structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303 Structure, Process, Technology, Policy, and Metrics 304 Structure (People and Agents) 305 Process 305 Technology 306 Metrics 306 Policy 306 Other Considerations 307 Structure of Agent Fleets 308 Fleets as the Scaling Unit of the Mesh 309 Dynamic Membership and Fluid Boundaries 310 Core Services as the Glue 310 Alignment to Missions, Not Just Functions 311 Key Roles in Fleet Management 311 Organizational Patterns for Fleets 312 Structure of Agent Teams 313 Agent Owner 315 Agent Engineers 315 Reliability and Operations Specialists (Agent SREs) 316 Governance and Certification Lead 316 Evaluation and Human-in-the-Loop Supervisor 316 Policy and Ethics Liaison 317 Release Manager 317 Transition Considerations 318 Human Impact and Ethical Foundations 319 Communication, Trust, and Cultural Adaptation 319 Structured Transition Through Reskilling, Support, and Governance 320 The Future of Work 322 Jobs: From Automation to Autonomy 323 Uneven Impacts and Workforce Polarization 325 Emergence of Hybrid Professions and Operating Models 326 Human Purpose, Adaptability, and Continuous Learning 327 Summary 328 14. Agent Factory: Building Agents at Scale. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 329 Agent Development Cycle 330 Building Agents at Scale 332 Fleets 333 Fleet Organization 335 Building Fleets 339 xii | Table of Contents

Operating Agents at Scale 340 Deploying Fleets 340 Monitoring Fleets: Fleet Observer Agents 341 Updating and Retiring Fleets 342 A More Distant Future 343 Agents Building Agents 343 Larger Abstractions 344 Summary 345 15. A Practical Roadmap for Implementation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347 Strategic Foundations 349 Phase 1: Formulate Strategy 349 Phase 2: Design Architecture 350 Phase 3: Identify Candidate Pipeline 350 Phase 4: Select MVP 351 Technology Build / Industrialization 351 Build Technology Foundation 352 Industrialize Technology Foundation 354 Secure Technology Foundation 354 Manage Models and Operations 355 Agent and Fleet Factories 356 Build Enterprise-Grade Agent Framework 358 Build Enterprise-Grade Agent Fleet/Ecosystem Framework 360 Establish Agent/Fleet DevSecOps 361 Create Agent Factory 363 Create Fleet Factory 364 Organizational and Operating Model 366 Establish New Operating Model 367 Manage Change 368 Train Staff and Build Skills 369 Governance and Certification 369 Establish Agent Governance and Certification 371 Establish Fleet Governance and Certification 371 Summary 372 Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375 Table of Contents | xiii

Foreword We find ourselves at an unusual moment in the evolution of artificial intelligence. The excitement is real, the investments are enormous, and the technical break‐ throughs are astonishing. Yet inside most companies, the story is quieter and more complicated. Many AI initiatives remain stuck in what people jokingly call “proof-of- concept (or POC) purgatory.” These initiatives look impressive in a demo, maybe even inspiring, but they rarely make it into the production systems that run the busi‐ ness. Teams are experimenting, executives are searching for return on investment, and somewhere between those two realities the work tends to stall. This book arrives at exactly the right time. It shines a light on the gap between early promise and operational reality. And it answers a question that almost every enterprise is wrestling with: “What does it actually take to run agents in the real world at scale?” I first crossed paths with Eric through an article he wrote comparing several agentic frameworks. The industry had been drowning in surface-level feature lists, but his work was different. He approached the problem like a scientist. Instead of asking what features a framework offered, he asked whether it could be trusted to operate inside an enterprise. He evaluated reliability, traceability, observability, and explaina‐ bility. He proposed a grading system that held each framework to the standards that real production environments demand. I contacted him as soon as I finished reading it. That conversation eventually became a collaboration on the idea of the enterprise agent ecosystem, some of which forms the backbone of this book. My own perspective comes from a career spent trying to bridge the distance between intelligent systems and the infrastructure they depend on. I began in research during my PhD and postdoc years. After that, I founded a company, then moved to Google to work on conversational systems, and today I lead AI efforts at Confluent. Across all of these environments, I have spoken with hundreds of teams trying to get value out of AI. The same theme appears again and again: isolation is the enemy of scale. A clever model or a clever agent is not enough. The real challenge is everything around it. xv

There are many books that teach you how to build an individual agent. How to write a prompt. How to attach a tool. How to get the model to plan and reason. This book is about what happens next. Once you build one agent, you will inevitably build many. And once you have many, the work stops being about the intelligence of a single component and starts being about the ecosystem that surrounds it. If the industry forecasts are even directionally correct, future enterprises will operate thousands or even millions of these agents. The question is no longer whether we can build an agent. It is how we manage a population of them. To answer that, this book reframes whether to consider deployable agents at all. Early prototypes were often fragile, opaque experiments. They worked until they did not. When they failed, no one could explain why. When they succeeded, no one could guarantee the result would repeat. Moving from experiments to production requires something sturdier. The authors argue that an enterprise-grade agent must be discoverable, because in a large ecosystem, the right component must be easy to find. It must be observable and traceable, because operators need to see the chain of reasoning and the steps an agent took. It must be reliable and explainable, because unpredictable behavior undermines trust and makes real deployment impossible. This leads naturally to the microagent model described in the book. Instead of treat‐ ing an agent as a monolithic script wrapped around a large language model (LLM), the authors treat it as a microservice. It has a container. It has interfaces. It has a set of operational guarantees. This shift allows agents to inherit decades of engineering wisdom: clean deployment pipelines, isolation, fault tolerance, reproducibility, and the security patterns that enterprises already expect. But the authors don’t stop there, and this is where the book truly stands apart. Indi‐ vidual competence is necessary but not sufficient. Once agents must work together, the system begins to resemble a distributed organization. Communication patterns matter. Coordination matters. Discovery matters. Governance matters. Without structure, the ecosystem collapses under its own complexity. This is where the concept of the agentic mesh becomes essential. The book describes an ecosystem where agents can find one another, exchange context, coordinate long-running tasks, and act with accountability. And the authors explain why this mesh must be event-driven. Traditional request-response APIs assume that both sides are ready at the same moment, but agents do not behave this way. They start, pause, wait for input, delegate work to other agents, and resume hours later. Their conversations are not synchronous calls. They are living, ongoing interactions. The authors argue that event-driven communication is the only pattern capable of supporting this behavior at scale. Events allow decoupling. They allow resilience. They allow many agents to observe the same state change and react xvi | Foreword

independently. In other words, events allow an ecosystem to emerge rather than a brittle network of point-to-point calls. For someone who has spent years working in data streaming and real-time architec‐ tures, this argument resonated immediately. If agents are to become the nervous system of the enterprise, then a streaming substrate is the circulatory system that keeps them informed and connected. The book shows in clear detail how streaming patterns align with the shape of agent interactions and why they are foundational for large-scale coordination. There is one more dimension the authors explore that is perhaps the most impor‐ tant of all: trust. It is easy to talk about trust in abstract, emotional terms, but in practice, trust is a set of engineering requirements. The book introduces a layered model that treats trust as something that spans the entire system. At the bottom are identity, authentication, and authorization. Above that are policy, certification, and governance. Each layer is concrete. Each layer sets boundaries. Each layer creates accountability. This structure ensures that an agent is not only secure in how it connects but is also certified to behave in alignment with organizational rules and expectations. This book provides a blueprint for the next phase of AI. It moves the conversation from the intelligence of the model to the architecture that allows intelligence to function at scale. It acknowledges that the future will not be shaped by isolated experiments but by interconnected systems that behave predictably even as they act autonomously. And it offers the vocabulary and the structure needed for enterprises to move from prototypes to production. In a moment when the industry is searching for clarity, Agentic Mesh offers some‐ thing rare: a grounded path forward. — Sean Falconer Head of AI, Confluent Foreword | xvii