Multi-agent systems get expensive and chaotic when every agent receives the full conversation, full tool history, and full memory state. In this session, we’ll look at how to design leaner AI systems on Azure using Azure OpenAI with Microsoft Agent Framework or LangGraph, focusing on the architectural patterns that reduce context bloat without losing capability. We’ll cover scoped context, targeted handoffs, short-term versus long-term memory, planner and router patterns, and techniques for keeping each agent focused on only the information it actually needs. Attendees will leave with practical design patterns for building multi-agent systems that are faster, cheaper, easier to debug, and easier to scale.

You will learn:

• About performance impacts of context size
• About leveraging long-term memory
• How to reduce token costs