Strip away the hype and an AI agent is something an architect can reason about: a harness, an LLM, and a directive. That's the starting point. From there, everything we already know about distributed systems comes rushing back.

This session builds the architecture of agents from first principles. We start with that minimal definition and grow it—agents fronted by chat interfaces, ephemeral agents triggered by queue messages and data changes, and harnesses that reach outward by invoking CLIs, MCP servers, and other tools. We'll look at agents that hold A2A conversations with peer agents, and confront the questions that follow: security, scalability, and reliability at scale.

Along the way a useful lens emerges—an agent looks a lot like a microservice, and the disciplines that tamed microservices apply here too. We'll map the emerging standards (MCP, A2A, and more) and the supporting cast of tool and agent discoverability, registries, and governance.

Come away with a mental model and a vocabulary for the three days ahead.

Learning points:

• A working definition of an AI agent and how to extend it. Attendees will leave able to describe an agent as a harness, LLM, and directive—and reason about how that core scales into chat-driven, event-driven, and tool-invoking variations.
• How proven distributed-systems discipline applies to agents. Treating an agent like a microservice gives architects a ready framework for security, scalability, and reliability, rather than inventing agent architecture from scratch.
• The emerging standards and infrastructure that make agents interoperate. A practical map of MCP, A2A, and the discoverability, registry, and governance pieces needed to build agents that find and work with each other.