Everyone is talking about AI agents. Very few people have actually built one that works in a real enterprise environment — with proper authentication, governed data access, and human approval gates before anything irreversible happens.
This session changes that. In 60 minutes, we build a production-ready AI agent from scratch on Microsoft Foundry — live, with code, on stage — starting from an empty project and ending with a deployed agent that can answer questions over enterprise data, call external tools via MCP, and route uncertain decisions to a human for approval.
We'll cover the five things every first-time Foundry agent builder needs to understand:
Step 1 — Agent anatomy: what a Foundry agent actually consists of — the system prompt, the tools, the memory, the reasoning loop — explained without jargon so you can explain it to your manager too.
Step 2 — Connecting your data: how to ground your agent on real enterprise data using Foundry IQ instead of static documents — so it always answers from your latest information, not last month's PDF.
Step 3 — Wiring tools via MCP: how to give your agent the ability to take actions — calling APIs, querying databases, sending notifications — using the Model Context Protocol so tools are securely registered and scoped.
Step 4 — Securing with Entra: how to wire Entra Agent Identity so your agent has its own identity, its own permissions, and can only access data and tools it's explicitly authorized for.
Step 5 — Human-in-the-loop: how to add a Task Adherence approval gate so your agent pauses and asks a human before taking any action that can't be undone.
By the end of the session, you'll have seen a complete, working Foundry agent built live — and you'll know exactly how to build your own. GitHub repo included.