Microsoft Foundry + Agent Framework: deployment tutorial

This walkthrough follows a sales meeting preparation agent built in Python with Microsoft Agent Framework. The developer starts with code running locally, then deploys it to Microsoft Foundry Agent Service through the AI Toolkit extension in Visual Studio Code. The production example uses Microsoft's hosted service; the local development step does not establish offline operation or local model inference.

The presenter explains that deployment assigns an identity through Microsoft Entra and that each user session gets an isolated microVM. Agents can use their own scoped permissions or act on behalf of a signed-in user. The demonstration also covers organizational guardrails, conversation traces and publishing to Microsoft 365 and Teams.

The sales agent uses Work IQ for workplace context, Fabric IQ for business operations data and Foundry IQ for organizational knowledge. It produces a Word briefing and a PowerPoint presentation for a customer meeting. GitHub Copilot SDK drives the AI agent's tool loop, while Microsoft Agent Framework supplies middleware that blocks document generation when required grounding data is missing. This is a demonstrated quality control, not proof that outputs cannot contain errors.

The tutorial also shows Azure Developer CLI templates for scaffolding a project and local debugging. It ends with fleet monitoring in the Foundry portal, including estimated cost, token usage, run volumes and success rates.