AI Flow Builder
AI Flow Builder turns a plain-English description into a complete, working Creator Studio flow: conversation cards, API definitions, an authenticated API Server, scripts, and error handling.
This section walks through the full lifecycle with one real example - letting employees add a user to a Microsoft Entra security group from the virtual agent, powered by the Microsoft Graph API. The same pattern applies to any flow you build.
What you will see
- Generate the flow with AI Flow Builder, including the API Server it creates for you along the way.
- Test the APIs as AI Flow Builder asks, then re-test any API later with the Test API button.
- Deploy the flow to Production and mark it as a top-level flow.
- Create and publish a Trigger so users can launch the flow with natural phrases.
- Verify the result live in Microsoft Teams, from trigger phrase to success card.
Why this example
Adding a user to a security group exercises every interesting part of the platform: OAuth 2.0 client-credentials auth, three different Graph endpoints (user search, group search, add member), dynamic dropdowns built from API responses, a confirmation step, and a write operation you can verify in your tenant.
Prerequisites
Before you start, have the following ready:
- Creator Studio access under Virtual Agent Management, with the beta AI Flow Builder enabled and an editable working folder (My folder).
- A Microsoft Entra app registration with application (not delegated) Microsoft Graph permissions:
User.Read.All,Group.Read.All, andGroupMember.ReadWrite.All(orGroup.ReadWrite.All), with admin consent granted. - Your app's connection values: Client ID, Client Secret, OAuth 2.0 token URL (
https://login.microsoftonline.com/<your-tenant-id>/oauth2/v2.0/token), scopehttps://graph.microsoft.com/.default, and base URLhttps://graph.microsoft.com/v1.0. - A target channel where your virtual agent is published - Microsoft Teams in this guide.
Treat the Client Secret like a password. Enter it only in Creator Studio's credential fields (the UI masks it), never in flow text, sample queries, or documentation. Screenshots in this guide redact the demo tenant's identifiers.
What's next
- Generating the flow - the full AI Flow Builder walkthrough, from prompt to saved components.
- The generated flow and its components - what AI Flow Builder produced and how to inspect it.
- Deploying and publishing - mark the flow top-level, deploy to Production, create a Trigger.
- Testing end-to-end - run the trigger phrase in Teams and verify the live result.
- Troubleshooting and security notes - common failure modes and the security-hardening checklist.