Listed under Roadmap ID 575282, Microsoft is rolling out a capability for Microsoft Contact Center AI (CCAI) that allows organizations to connect existing agents from Azure Foundry to agents operating inside CCAI. Rather than forcing developers and administrators to recreate logic, prompts, or workflows from scratch inside a separate contact-center toolset, the new Contact Center AI Conversational Agent Builder lets a CCAI agent treat an existing Foundry agent as a connected agent.
How the CCAI-to-Foundry Connection Works
According to Microsoft’s roadmap entry for ID 575282, the integration centers on query routing and flexible response delivery:
- Builder Integration: Administrators and makers working inside the web-based Contact Center AI Conversational Agent Builder can link existing agents created in Azure Foundry as connected agents for their CCAI agents.
- Query Delegation: When a customer interacts with the contact center, the primary CCAI agent acts as the orchestrator and front-end interface, routing relevant queries to the connected Foundry agent.
- Multimodal Output: Once the connected Foundry agent returns an answer, the CCAI agent responds back to the end user in either plain text or spoken audio delivered via text-to-speech (TTS).
It is worth drawing a clear technical boundary around that last point. Based on the roadmap description, the speech capability described for this feature relies on text-to-speech conversion at the CCAI response layer. It does not establish that an arbitrary connected Foundry agent is natively processing live audio streams end-to-end on its own, even though Microsoft Learn separately documents dedicated Foundry voice agent previews and Voice Live API accelerators for telephony workloads.
Release Timeline and Cloud Scope
For IT planners tracking deployment windows, Microsoft’s AI at Work roadmap provides the following parameters for Roadmap ID 575282:
| Attribute | Roadmap Specification |
|---|---|
| Product | Microsoft Contact Center AI |
| Platform | Web |
| Status | Rolling out |
| Preview Availability | October 2026 |
| General Availability | November 2026 |
| Release Rings | Preview, General Availability |
| Cloud Instance | Worldwide (Standard Multi-Tenant) |
As with any entry on the AI at Work roadmap, Microsoft explicitly cautions that roadmap dates and descriptions are estimates subject to change, and that "Rolling out" indicates an update is beginning to reach eligible tenants rather than arriving everywhere simultaneously. Notably, the listing only specifies Worldwide (Standard Multi-Tenant) environments, with no timeline currently listed for GCC, GCC High, DoD, or sovereign cloud instances.
Where This Fits in Microsoft’s Agent Architecture
To understand why this matters to enterprise architects, it helps to look at how Microsoft currently divides its agent ecosystem. In Microsoft Learn’s documentation for its Agents Hub and Foundry Agent Service, Microsoft positions Microsoft Foundry as a pro-code "AI app and agent factory" where developers build, customize, and govern custom agents—complete with first-class identities via Microsoft Entra Agent ID and centralized oversight in the Foundry Control Plane. Meanwhile, Foundry Agent Service already supports publishing agents into surfaces like Microsoft Teams and Microsoft 365 Copilot, and Dynamics 365 Contact Center supports integrating Azure bots across messaging and omnichannel pipelines.
Bringing existing Foundry agents directly into CCAI’s Conversational Agent Builder extends that "build once, surface anywhere" philosophy to customer service operations. In practical terms, if an enterprise engineering team has already built and vetted a back-end Foundry agent that troubleshoots hardware SKUs or checks order logistics, the contact center team can wire that agent behind a customer-facing CCAI agent instead of maintaining duplicate prompt engineering and tool definitions in two places.
Practical Questions Architects Should Check During Preview
While the roadmap entry confirms the core routing and text-to-speech flow, it stops short of documenting the underlying plumbing. Before moving a connected CCAI-and-Foundry workflow into production, enterprise administrators should verify several operational details as Microsoft publishes full implementation documentation:
- Identity and Resource Permissions: In Microsoft Learn’s existing guidance on connected agents in Microsoft Foundry, Microsoft warns that shared development identity permissions do not automatically transfer when agents are published or connected, requiring administrators to reconfigure permissions for downstream Azure resources. Admins should verify how CCAI authenticates to a connected Foundry agent and what RBAC roles are required.
- Metering and Consumption Costs: According to Microsoft Learn’s Foundry Agent Service documentation, running Foundry agents can incur separate Azure consumption charges across generative AI token input and output, speech synthesis, Application Insights logging, and connected data tools. Teams should clarify how queries routed from CCAI are billed across their Microsoft 365/CCAI licensing and Azure subscriptions.
- Fallback and Latency Behavior: Chaining one agent to another introduces an extra network and reasoning hop. Particularly when the CCAI agent converts the returned answer into audio via text-to-speech, voice-channel administrators will want to test round-trip latency and configure clear fallback behavior in case the downstream Foundry agent times out or fails to resolve a query.
Summary
Roadmap ID 575282 is a focused, practical quality-of-life upgrade for organizations investing in both Azure Foundry and Microsoft Contact Center AI. By allowing CCAI’s Conversational Agent Builder to route queries to existing Foundry agents and deliver answers via text or text-to-speech, Microsoft is reducing duplicate development work between back-end AI engineers and front-line contact center teams. With preview rollout slated for October 2026 and general availability targeted for November 2026 in Worldwide multi-tenant clouds, administrators should watch their CCAI web portals and test permission boundaries and voice latency before lighting up customer-facing queues.