From prototype to production: build and run agents at scale | BRK241

Tina Schuchman and Jeff Hollan present a Microsoft Build 2026 breakout on taking AI agents from a prototype to a production-ready, scalable deployment using Azure AI Foundry Agent Service (Foundry Agent Service) and Microsoft Agent Framework.

Overview

The session focuses on what changes when an agent moves from a local demo to an enterprise deployment: identity and authentication, secure networking and data protection guardrails, evaluation strategies, and ongoing lifecycle management. It also highlights how coding agents such as GitHub Copilot can integrate into the developer workflow.

What the session covers

Enterprise AI platform context

• Introduction to Microsoft Foundry as an “enterprise AI operating system” concept for building and operating agents.

End-to-end lifecycle: build, deploy, operate

• A demo-driven walkthrough structured around three phases:
  • Build
  • Deploy
  • Operate

Identity, authentication, and guardrails

• Managing authentication for the agent.
• Setting data protection guardrails as part of making the agent production-ready.

Voice enablement

• Enabling voice for a Foundry-deployed agent to support hands-free interaction.

Frameworks and tooling

• Overview of Microsoft Agent Framework 1.0.
• Foundry Toolkit GA (general availability) callout.

Approval-driven workflows

• An approval-driven workflow model.
• Automatic session resumption as part of the agent experience.

Publishing and integrations

• Publishing agents to Microsoft Teams.
• Copilot integration mentioned as part of the distribution/integration story.
• GitHub Copilot is referenced as a coding agent that integrates directly into the workflow.

Evaluations

• Selecting and customizing evaluation logic.
• Using built-in evaluators and creating custom evaluators.

Resources

• https://aka.ms/build26/BRK241
• https://aka.ms/build/foundrydiscord

Speakers

• Tina Schuchman
• Jeff Hollan