I'm setting up Agent-to-Agent (A2A) orchestration in Azure AI Foundry (New) and both of my published Agent Applications have their identity provisioning permanently stuck at "Creating".
Environment
- Region: East US 2
- API Version:
2025-10-01-preview
Problem
When I query the Agent Application resources via ARM API (GET .../applications/{name}?api-version=2025-10-01-preview), the top-level provisioningState shows "Succeeded", but the nested identity states are stuck:
"agentIdentityBlueprint": { "provisioningState": "Creating" },
"defaultInstanceIdentity": { "provisioningState": "Creating" }
The Entra ID service principals do exist and are enabled — the identities were created, but the provisioning state never transitioned to "Succeeded".
Impact
This blocks A2A tool calls. When my orchestrator agent tries to invoke a sub-agent via the a2a_preview tool, I get:
Error code: tool_user_error
Error message: 400 Failed to fetch agent card: Response status code does not indicate success: 401 (PermissionDenied)
I've verified:
- RBAC is correct: Azure AI User role assigned on the Agent Application resources for both the orchestrator's agentic identity and the project managed identity
- Direct endpoint calls work: Calling the Agent Application endpoint with a user bearer token returns 200 OK
- Auth type doesn't matter: Tested with both
AgenticIdentityToken and ProjectManagedIdentity connection types — same 401
- Cannot modify or delete the stuck resources: PUT and DELETE operations return
SystemError from managementfrontend in eastus2
Additional details
- Two separate Agent Applications are affected, suggesting this is systemic in the region rather than resource-specific
Questions
- Is there a known issue with Agent Application identity provisioning in East US 2?
- Is there a way to re-trigger or unstick the identity provisioning?
- Would creating the Foundry resource in a different region (e.g., Sweden Central) avoid this issue?