This session is for anyone building agents that need to do more than just answer questions. Using the AI Toolkit in VSCode, I’ll walk through how to go from an idea to a working agent that can reason, take action, and connect to tools and data sources. Everything runs inside VSCode, so you stay in your flow without switching environments.

We’ll look at how to use models from Azure AI Foundry, but also how to use local models from Foundry local and open-source models. Switching between them is straightforward, and I’ll show what to pick depending on the context.

The AI Toolkit supports grounding with files, URLs, and structured data, and tool calling. You can inspect traces, test different model configurations, and use the playground to iterate quickly. It also includes support for agent memory and deploying agents to Azure directly from VSCode.

When it’s time to move from prototype to production, the Microsoft Agent Framework comes in. I’ll show how to take what you’ve built in the Toolkit and plug it into the framework to handle scale, reliability, and integration.

If you want to build agents that do more than just answer questions, this session should help. It’s hands-on, based on real examples, and focused on what’s possible right now. I’ll also cover what’s coming next.