Agentic Mesh Ecosystem Patterns with Eric Broda
AgileData Podcast #57
Join Shane Gibson as he chats with Eric Broda on the patterns required to create an ecosystem to support the use of Agents in enterprise organisations.
Listen
Listen on all good podcast hosts or over at:
https://podcast.agiledata.io/e/agiledata-57-agentic-mesh-ecosystem-patterns-with-eric-broda/
Read
Read the podcast transcript at:
https://agiledata.io/podcast/agiledata-podcast/agentic-mesh-ecosystem-patterns-with-eric-broda/#read
Google NoteBookLLM Briefing
Briefing Document: Agentic Mesh Ecosystems
Source: "Agentic Mesh Ecosystem Patterns with Eric Broda - AgileData.io" Podcast, January 24, 2025.
Introduction:
This podcast interview with Eric Broda explores the concept of the "Agentic Mesh," a novel approach to leveraging AI agents within enterprise organisations. Broda, a tech veteran with a background in API, service mesh, and data mesh implementation, now focuses on building ecosystems for enterprise-grade autonomous agents. This briefing summarises the core ideas and key takeaways from the discussion, focusing on definitions, patterns, and implementation considerations.
Key Definitions:
Generative AI (Gen AI): Broda defines Gen AI as a "superpower" enabled by Large Language Models (LLMs), allowing for natural language interaction and, potentially, reasoning by computers. He sees it evolving from content creation to a cornerstone of broader ecosystems. "It’s a superpower that lets computers be heck of a lot smarter than they have been in the past."
Agent: An agent utilises an LLM to plan and execute tasks using tools. A more sophisticated agent can learn from past interactions and create new capabilities. Importantly, an agent also interacts with its environment via tools, unlike a standard LLM interface. "An agent is, it uses an LLM, large language model. It can plan its activities, it can then execute those activities or tasks, and it can use tools to actually do that."
Agentic Mesh: The Agentic Mesh is defined as an ecosystem where autonomous agents can find each other and collaborate safely, interact, and transact. This ecosystem also needs to be "enterprise-grade". "Agentic mesh is an ecosystem that lets autonomous agents find each other and safely collaborate. Interact and transact." This ecosystem also considers the consumer, producer, agent and operator experiences.
Core Themes and Concepts:
Ecosystem Model (Mesh): The "mesh" is framed as an ecosystem, similar to platforms like Airbnb or Uber. It enables producers (those who create agents) and consumers (those who use agents) to find each other, interact, and transact. Broda emphasises the simplicity of this concept despite confusion created by some in the tech industry. "So for me, a mesh is just an ecosystem. And if I have a data mesh, it lets folks who want to consume data and folks who want to produce data, find each other, interact, collaborate, and transact."
The Five Planes: Broda outlines five key "experience planes" that make up the agentic mesh:
Consumer Plane: The consumer interacts with an agent, via a user-interface, or chat like interface where they are looking to achieve a task. This may initially resemble an "app store for agents", but Broda believes it will evolve towards a more intuitive, chat-based interface, possibly multimodal.
Producer Plane: This plane is for people who build agents. It provides the templates and toolkits needed for planning, execution, and use of tools as well as making agents enterprise grade. This plane also provides monitoring, version and upgrade capabilities.
Governance Plane: This provides policies for agents and for compliance to the organisation's requirements. Here agents and owners are provided the tools to demonstrate the agents are working as expected. Broda also uses the term "certification" instead of governance.
Agent Plane: This plane focuses on how agents find each other, interact, and collaborate. A "super planner" or "super orchestrator" is where a request initially comes into the agent ecosystem. It creates a plan based off of the agents available, then gives the tasks to each agent to execute.
Operator Plane: Provides the technology and platform required to operate the agents such as Kubernetes, cloud technologies and managing LLMs at scale.
Agentic vs. Human Work: Broda argues that while humans will remain in the loop (especially for governance and oversight), agents will increasingly replace the people who do a lot of work in the background. He foresees automation of business processes and a potential decrease in human intervention where those processes can be made repeatable. The key element to this is that current human led processes have many unstructured, untidy parts and this can be handled by AI agents. "My proposition is a lot of those people in the loop today, humans in the loop can be represented by agents."
Microservices Architecture: Broda emphasizes that agents should be treated as microservices – small, independent, and containerised entities. This approach facilitates integration with existing enterprise infrastructure, ensuring security, discoverability, and operational efficiency. "Every agent is, and we’ll talk about this, it’s a microservice. It’s in a container, it’s deployed, it has some endpoints, and it has a way of interacting. It has an LLM, so we have a smart, a very smart microservice."
Importance of Determinism and Repeatability: While LLMs are not 100% deterministic, Broda highlights techniques to improve reliability. He uses the term "repeatable" as an approach to build processes so you can expect a predictable outcome each time, despite the unstructured processes an agent can deal with. He suggests that well-defined policies, appropriate context, schema constraints, and prompt engineering can greatly increase the repeatability and reliability of agent behaviour. He argues the goal is not to achieve 100% deterministic behaviour, but to achieve significantly better outcomes than the current human systems.
Enterprise Grade Agents: Broda defines enterprise-grade as agents that fit into a normal enterprise operating environment and meet service level expectations around discoverability, observability, operability, and security. This can be achieved by implementing agents as microservices with OAuth2 and RBAC, logging and alerting capabilities. He stresses that the current tooling for creating AI agents is not "enterprise grade". "It means that they’re going to, simplistically, they’re going to fit into a regular, normal enterprise’s operating environment and meet the regular, normal, service level expectations that they have."
Challenges & Future Outlook:
Transitioning from POCs: Broda notes that most current AI projects are in the "science experiment" or "proof of concept" stage, often with limited real-world value due to the lack of enterprise-grade agent toolkits. He thinks there will be a shift from this to the adoption of enterprise-grade agents and tools that will accelerate development and adoption.
Governance and Policies: The interview highlights the challenge of implementing robust governance and policies for autonomous agents. It's not a solved problem. Broda sees a need for federated ownership and robust certification mechanisms to ensure agents operate within defined boundaries.
The Agent Plane: The agent plane presents a challenge due to the complexity of building agents that can make independent decisions and work recursively. It needs optimization and contract patterns for it to work effectively.
The Next Gold Rush: Broda states there has been billions of dollars of investment recently by all the major tech companies into the "agentic future." and now is the time to prepare, to stake a claim in this new area.
Conclusion:
Eric Broda's perspective on the Agentic Mesh offers a thought-provoking vision of the future of enterprise AI. His emphasis on ecosystems, microservices, and "enterprise-grade" capabilities provides a practical framework for transitioning from experimental AI projects to real-world business value. While challenges remain, particularly in the areas of governance and agentic interactions, the potential benefits of an agent-based architecture are substantial, making this an area worth close attention.
Key Takeaway: It's about moving from "Ask AI" or basic chatbots to creating autonomous agents that work in an enterprise setting using a microservices approach, with emphasis on the ecosystems, observability and discoverability.