Microsoft Events presents a panel discussion at Ignite 2025, featuring Mark Austin, Luis Héctor Chávez, Adam Orentlicher, and Naga Surendran as they share real-world approaches to architecting AI agents using Azure and open AI frameworks.

Panel: Real-World Architectures and Lessons from Scaling AI Agents on Azure

Speakers: Mark Austin, Luis Héctor Chávez, Adam Orentlicher, Naga Surendran
Event: Microsoft Ignite 2025 – Session BRK114

Overview

This session provides an in-depth look at how enterprise teams are building and deploying AI agent solutions using open frameworks and Azure services. The panel explores:

  • Architectural fundamentals for scalable agent solutions
  • Model selection and evaluation strategies (including Microsoft Foundry)
  • Production readiness: orchestration, reliability, trust, and security
  • Human-in-the-loop systems and governance for responsible AI
  • Autoscaling, fault tolerance, and observability in real deployments
  • Best practices and key advice for AI adoption and iterative prototyping

Key Session Topics

Scaling AI Agents Responsibly

  • Strategies for architecting scalable, reliable agentic systems
  • Balancing platform reuse with customization for business impact
  • Ensuring governance and trust throughout the lifecycle

Human-in-the-Loop AI Design

  • Integrating expert oversight in training and production
  • Designing systems for continuous feedback and improvement

From Prototypes to Production

  • Emphasis on live demos and interactive models over static presentations
  • Iterative prototyping to accelerate learning and mitigate risk

Model Choices and Multi-Model Approaches

  • Using Microsoft Foundry to evaluate and select the right models for each use case
  • Multi-model strategies for matching requirements to capabilities

Architecture for Resilience and Observability

  • Building systems with autoscaling and fault tolerance
  • Handling GPU scaling and planning for model failover
  • Observability practices for trustworthy operations

Practical Lessons & Takeaways

  • Security and orchestration are ongoing production concerns
  • Measurable business outcomes require an iterative and outcome-driven approach
  • Teams should focus on platform capabilities, not reinventing the wheel

Resources

About the Speakers

Experienced customers and partners from diverse industries, including Mark Austin, Luis Héctor Chávez, Adam Orentlicher, and Naga Surendran, contribute their perspectives on successfully implementing agentic AI solutions with Azure.