AI-3026: Agent Development on Microsoft Foundry

Note: This lab pre-provisions Azure Foundry resources at start — allow up to 10 minutes for the environment to become ready before beginning the exercises.

Get hands-on with Microsoft Foundry by building your very first agent — no code required. You will provision a Foundry project, deploy a chat model, and build the Halcyon Insurance intake-triage agent directly in the Foundry portal, then attach the Foundry Toolkit VS Code extension so you can iterate on your agent from an editor. By the end of this lab you will be able to explain the Foundry account → project → agent resource model, build and test a prompt-driven agent, and connect a local VS Code environment to a live Foundry project.

Note: This lab pre-provisions Azure Foundry resources at start — allow up to 10 minutes for the environment to become ready before beginning the exercises.

Learn how to extend a Foundry agent with a custom Python function tool so it can reach real business data — the pattern that separates a chatbot from a working assistant. You'll register a typed Python function on a Foundry agent, watch the LLM decide when to call it, and observe how tool_choice controls tool selection. By the end of this lab you will be able to author function tools that a Foundry agent can call to look up business records, and reason about when to force a tool call versus let the model choose.

Note: This lab pre-provisions Azure Foundry resources at start — allow up to 10 minutes for the environment to become ready before beginning the exercises.

Model Context Protocol (MCP) is how portable, out-of-process tools become part of an agent's tool catalog. By the end of this lab you will be able to publish a Halcyon Insurance parts-and-repair-cost tool set as an MCP server and consume it from a Foundry agent — reusing the same server across multiple agents without redeploying either.

Note: This lab pre-provisions Azure Foundry resources at start — allow up to 10 minutes for the environment to become ready before beginning the exercises.

Ground an AI agent on your organization's document library using Foundry IQ, Microsoft Foundry's native retrieval-augmented-generation surface. By the end of this lab you will be able to bulk-index a policy document library into a Foundry IQ vector index, attach the index as a tool source on a Foundry agent, and compare grounded vs ungrounded agent responses to see how RAG eliminates hallucinations. You'll work in Halcyon Insurance's policy-lookup workflow — indexing four fictional Halcyon coverage documents and building an agent that cites specific policy numbers when it answers questions about deductibles and coverage limits.

Note: This lab pre-provisions Azure Foundry resources at start — allow up to 10 minutes for the environment to become ready before beginning the exercises.

Learn to build production AI agents with the Microsoft Agent Framework — the newer 2026 GA Python SDK that powers multi-agent orchestration and Agent-to-Agent (A2A) federation on Microsoft Foundry. You will use the FoundryChatClient and Agent primitives to build a Halcyon Insurance adjuster expense-reimbursement categorizer, then add streaming and multi-turn conversation state.

By the end of this lab you will be able to build a Microsoft Agent Framework agent that connects to a Foundry model, categorize inputs consistently across runs, stream token output live to a user, and preserve multi-turn conversation state.