Microsoft Events presents a deep dive session on multi-agent orchestration and automation in Microsoft Foundry, featuring Christof Gebhart, Shawn Henry, Tina Manghnani, and Mark Wallace. Learn about scaling from single agent prototypes to secure, governed, and observable fleets, leveraging Semantic Kernel, AutoGen, and Azure AI.

AI-Powered Automation and Multi-Agent Orchestration in Microsoft Foundry

Presented by: Christof Gebhart, Shawn Henry, Tina Manghnani, Mark Wallace
Event: Microsoft Ignite 2025, Session BRK197

Overview

This session explores the journey from single-agent prototypes to orchestrated, policy-governed multi-agent systems using Microsoft Foundry. Attendees will learn best practices and advanced strategies for architecting, deploying, and managing AI agents at scale with strong focus on governance, monitoring, and human-in-the-loop design.

Key Topics Covered

  • Foundry Agent Framework (Semantic Kernel + AutoGen):
    • Framework for rapidly building and managing AI agents with robust orchestration capabilities
    • Leverages existing AI development tools such as Semantic Kernel and AutoGen
  • Fleet-Level Orchestration:
    • Techniques for scaling agent deployments and coordinating multiple agents in an ecosystem
    • Integrating shared state and agent-to-agent (A2A) activity protocols
  • Human-in-the-Loop Automation:
    • Incorporating manual decision steps and validations in automated workflows
  • Cross-Framework Compatibility:
    • Integrating external agent frameworks like LangGraph and OpenAI Agents SDK for greater flexibility
  • Containerized, Governed, Observable Agents:
    • Deploying agents as containers for scalability and manageability
    • Utilizing OpenTelemetry (OpenTel) for observability and monitoring of agent activity
  • Security and Governance with Microsoft Purview:
    • Integration with Microsoft Purview enabling policy enforcement and data governance
    • Demonstration of Purview policy middleware for secure agent workflows
    • Techniques for detecting and handling sensitive data (e.g., credit cards, SSNs)
  • Development Tooling and Migration:
    • GitHub Copilot extensions for agent code migration and development boosts within Visual Studio
    • MCP toolchains for managing complex multi-agent environments
  • Practical Demonstrations:
    • Setting up AZD Templates and Environment for Foundry-based projects
    • Real-world examples including BMW’s multi-agent ecosystem

Session Structure

  • Introduction: Framing AI-powered automation and multi-agent trends
  • Impact Case: BMW’s large-scale deployment
  • Lessons Learned: Insights from Microsoft and its partners, frequent development challenges
  • Workflow Deep Dive: Integrating Foundry’s agent orchestration with shared state and framework plug-ins
  • Security & Governance: Data protection with Purview middleware and real-time sensitive data analysis
  • Tools Walkthrough: Using Copilot extensions, Visual Studio tools, and AZD for rapid adoption

Resources and Further Learning

Speakers

  • Christof Gebhart
  • Shawn Henry
  • Tina Manghnani
  • Mark Wallace

This session is recommended for AI developers, architects, and platform engineers aiming to securely orchestrate, observe, and govern fleets of AI agents in the Azure ecosystem.