Agentic Mesh

In a world of thousands or perhaps millions of agents, the question is not “how to build an agent” but rather “how to manage and scale an agent ecosystem”.

It is about how to scale a fleet of agents, how to understand what they are doing, how to establish trust with agents, and how to allow agents to find each other and safely collaborate, interact, and even transact.

That is what this book is about. Enjoy!

The new book, “Agentic Mesh”, is available now from O’Reilly, Amazon, and wherever great books are sold.

–

We began conceiving the foundational concepts and architecture around ecosystems described in this book a long time ago, probably well before the current incarnation of large language model (LLM)-based agents appeared.

In fact, we have been building ecosystems like those described in this book for APIs, whose ecosystem we call a service mesh, for data and data products in a data mesh ecosystem. In this book, we are covering how to do it for agents in an agent ecosystem, which we call agentic mesh.

But first, if Agentic Mesh is an ecosystem, then what is the definition of an ecosystem? We use a pretty simple definition: an ecosystem is a set of interconnected parts that interact and depend upon each other. In technology, ecosystems emerge when different components—such as services, data, or agents—are designed to work together.

It is the ecosystem which provides services that make it easy and safe for participants in the ecosystem to find each other and safely collaborate, interact, and even transact. In a service mesh, APIs are the ecosystem’s participants, enabling services to discover, communicate, and collaborate reliably. In a data mesh, data products play this role, serving as standardized, trustworthy units of data that can be shared and reused across teams.

Now, why the interest in ecosystems? Today, we see innovations like OpenClaw, now with over 200,000 GitHub stars, clearly taking the developer world by storm. Thousands of OpenClaw agents are running, but perhaps the most interesting part about OpenClaw is its ability to work with other agents, not just on your own computing environment, but with other OpenClaw agents across the internet. The agent ecosystem has arrived. But days are early and there is still work to do to get OpenClaw and agent toolkits like it to be safe.

And, we also see industry leaders offering a compelling agent ecosystem vision: Jensen Huang (CEO, NVIDIA) says that “enterprises will have “a couple of hundred million digital agents, intelligent agents”; Satya Nadella (CEO, Microsoft believes that “Agents will replace all software”; and Andy Jassy (CEO, Amazon) says that “there will be billions of these agents, across every company and in every imaginable field.”

So it is not just us that see a clear emergence of the agent ecosystem. In fact, we expect that soon agents themselves will be the core participants in every business process. And, in this case they strongly believe that agents must all be explicitly designed so they can work together as safe building blocks of within an enterprise agent ecosystem.

But what does it mean to have “safe” agents and a “safe” agent ecosystem?

First, the very definition of an ecosystem - where many participants collaborate - leads immediately to a need to address scale. How can thousands of agents, each individually an independent entity, plan work, execute work, and deliver consistent outcomes at-scale?

Second, if we expect agents to work at scale in an enterprise then they also need to adopt all the characteristics of other enterprise applications. Simply put, agents must be “enterprise-grade”: Agents are enterprise-grade when they are secure, trusted, reliable, operable, and observable.

But it is not just the agents, rather the entire agent ecosystem must also be enterprise-grade! We need to be able to answer challenging questions like: How do agents gain an identity and have roles and permissions such that they can do real work in an enterprise? What services must a communication fabric provide to allow agents to collaborate, interact, and even transact safely? How can we understand what and agent has done? Why did an agent do what it did?

These are not just agent, but also agent ecosystem concerns, and that is what agentic mesh, and hence, this book, is about.

In this book we address enterprise-grade concerns head-on. But we recognize that every firm starts somewhere different, and your first agent ecosystem may start with a very small number of agents. But as they say, begin with the end in mind, and so hence we offer this book as guidance to plan ahead and design your agent ecosystem for growth, safety, trust, and scale.

–

The Early Reviews are In!

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

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. Eric and Davis Broda’s new book, Agentic Mesh, is a masterclass for how companies can build a ‘connective tissue’ for AI. Essential reading!”

Bruno Aziza, GVP Enterprise Software, IBM

Agentic architectures are on the rise and 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 the enterprise architecture blueprint we’ve been missing. Davis and Eric 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, Inductee, National Inventors Hall of Fame

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 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 & AI Strategist, Actian

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 & AI Strategist, former Accenture Chief Data Architect and R&D Lead

What You Will Learn

Readers of this book will learn how to design and govern large-scale agent ecosystems and their agents that are enterprise-ready:

• Foundations: What agents are, why ecosystems matter, and how to ensure security, observability, and explainability.

• Technology: How to build the mesh’s core plumbing—data, messaging, APIs, and models—with resilience, scalability, and zero-trust security.

• Agent, fleet, and factories: How to standardize templates, SDKs, and connectors, and scale fleets with orchestration patterns and DevSecOps.

• Organizations: How roles, processes, and culture evolve to integrate agents as trusted team members.

• Trust and Governance: How to certify agents and fleets, manage systemic risks, and balance central rules with delegated ownership.

• Roadmap and Strategy: How to align the mesh with enterprise goals, deliver MVPs, and build credible adoption roadmaps.

Book Chapters

The book is organized into three parts; Part 1 defines essential concepts and definitions; Part 2 defines the agentic mesh and its core components; and Part 3 explains how you can build your agentic mesh.

Part 1: Defining the Essentials

This section introduces readers to the core concepts in agentic mesh. Part 1 begins with the essentials—what agentic mesh is, why it matters, and how it fits into the broader story of AI and automation. From there, the book situates agents within history, distinguishing their unique role from earlier AI approaches and showing how they connect to the evolution of workflows. These early chapters also explain what makes agents different from generic AI models, grounding the discussion in simple definitions, key concepts, and practical explanations that are accessible to a wide audience.

Chapter 1: Agentic Mesh - The Essentials

Introduces agents - what are they, what do they do - and provides a concise definition of an agent and its capabilities that anchors all other chapters in the book

Chapter 2: Agentic Past, Present, and Future

Explains the agentic past, present, and (potential) future. Today’s agents stand on the shoulders of artificial intelligence (AI) and machine learning (ML) giants - learning from the past informs not only how we got here, but where we are going.

Chapter 3: Agents versus AI Workflows

Lots has been written about AI workflows, but how are today’s workflows similar to and different from agents? What we explain in this chapter is how they are intertwined and how they build upon each other.

Chapter 4: Agent Basics

Explains the core agent capabilities - what do they do, how do they do it, and what are the key requirements that set the stage for subsequent chapters on the agent and agent ecosystem architecture

Part 2: Defining the Agent Ecosystem: Agentic Mesh

The second part moves from foundational concepts to deep architecture and governance. It begins with agent design and enterprise-grade requirements, then builds upward into the mesh itself as an ecosystem of agents. Readers will see how architecture, registries, interaction management, and user experience form the scaffolding of agentic mesh, ensuring that agents are discoverable, observable, and able to collaborate at scale. Alongside these technical elements, these middle chapters address the user experience dimension, explaining how both people and agents interact within the mesh in transparent and predictable ways.

This part also introduces the critical governance and trust frameworks that ensure enterprise adoption. Topics like security, compliance, and systemic oversight are not treated as afterthoughts but as integral parts of ecosystem design. You will learn how to balance autonomy with control—enabling agents to act independently while embedding policies that maintain enterprise trust and regulatory readiness. By the end of Part 2, you will understand the full blueprint of agentic mesh—not just the pieces that make it run, but the safeguards that make it safe to adopt in complex, real-world environments.

Chapter 5: Agent Architecture

Defines the agent architecture. We specify the core principles and internal building blocks—LLM-driven task planning and execution, memory and context engineering, tool integration, messaging/workspaces, state management, and reusable agent types and patterns—needed to operate autonomous agents reliably at scale.

Chapter 6: Enterprise-Grade Agents

Defines enterprise-grade agents as MicroAgents—microservice-deployed, security-first, observable, and operable components that convert ambitious “millions of agents” visions into production reality by constraining autonomy inside disciplined engineering controls for reliability, scalability, explainability, discovery, and lifecycle management.

Chapter 7: Agentic Mesh is the Enterprise-Grade Agent Ecosystem

Presents Agentic Mesh as the enterprise-grade control plane that turns thousands of autonomous agents into a governed system by providing discovery and metadata through a registry, end-to-end traceability through monitoring, secure interaction APIs, and a marketplace and workbenches that operationalize lifecycle, policy, and compliance at ecosystem scale.

Chapter 8: Agentic Mesh User Experience (UX)

Introduces the Agentic Mesh UX as the enterprise adoption layer that makes a large, headless-agent ecosystem usable and governable by providing a single sign-on entrypoint, orientation and navigation, a two-sided marketplace for discovery and evaluation, and role-specific workbenches that translate trust, lifecycle discipline, and operational control into repeatable workflows.

Chapter 9: Agentic Mesh Registry

Frames the Agentic Mesh Registry as the system-of-record that makes the UX and the broader mesh operable by providing authoritative, machine-readable metadata for agents, users, conversations, interactions, workspaces, and policies, enabling reliable discovery, consistent policy enforcement, traceability, and durable coordination across long-running workflows.

Chapter 10: Agentic Mesh Interaction Management

Defines interaction management as the mesh’s runtime communication fabric—conversations and interaction lifecycles correlated by IDs and enforced by policy, delivered over an event-driven queue/pub-sub layer with replay—so users and agents can initiate, coordinate, observe, and recover work reliably at enterprise scale.

Chapter 11: Agentic Mesh Security Considerations

Explains how to secure the mesh by hardening transport with mTLS, enforcing strong identity and least-privilege authorization for both humans and agents, isolating and rotating secrets, treating the LLM as a first-class attack surface (prompt injection and jailbreaking), and closing the loop with behavioral monitoring plus containment and recovery so compromise can be detected, bounded, and remediated without losing auditability or operational continuity.

Chapter 12: Agentic Mesh Trust Framework and Governance

Defines a layered trust and governance model that makes agents deployable in production by tying each agent’s identity, declared purpose and policies, and enforced least-privilege access to explainable planning, end-to-end observability, repeatable certification, and lifecycle governance that keeps trust valid as agents change and scale.

Part 3: Building Your Agentic Mesh

The final part turns theory into action, guiding readers through how to build and scale an agentic mesh inside an enterprise. It starts with the operating model and team structures, showing how roles like Agent Owner, Fleet Manager, and Governance Lead emerge to support hybrid human–agent organizations. From there, readers are introduced to the agent factory, a repeatable way to design, certify, and scale agents while embedding governance by default. This ensures that agents are not just experimental prototypes but production-ready services that align with enterprise standards.

The book concludes with a practical roadmap and schedule, providing actionable steps for adoption. This roadmap helps you assess where you are today, what short-term wins are achievable for your organization, and how to progress toward a mature agentic mesh over time. By the end of Part 3, you will be equipped with the strategies, tools, and organizational practices needed to make agentic mesh real—not just as an idea, but as a working system inside your enterprise. This closing section ties the book together by turning vision into execution.

Chapter 13: Operating Model and Team Structure

Explores how organizations must adapt as agents become part of everyday operations. It discusses new roles, workflows, and accountability models required to manage thousands of autonomous agents.

Chapter 14: Building Agents at-Scale

Describes how to industrialize the process of creating, testing, and deploying agents. Just as manufacturing transformed production through standardization and automation, an agent factory introduces templates, pipelines, and quality controls for agent development.

Chapter 15: A Practical Roadmap

Ties everything together by providing a step-by-step guide for establishing an enterprise’s agentic mesh. It details the streams of work involved—technological, organizational, and governance—and explains how to align them under a cohesive implementation plan.

Thank You (From the Authors)

–

Looking for more?
👉 Discover the full O’Reilly Agentic Mesh book by Eric Broda and Davis Broda

🎧 Follow co-hosts John Miller and Eric Broda on The Agentic Mesh Podcast on Youtube, Spotify and Apple Podcasts. A new video every week!