2.2.3. Designing Multi-Agent Solutions Across Platforms
Most enterprise AI solutions aren't single-agent deployments — they involve multiple agents across multiple Microsoft platforms working together. The architect must design which agents live where, how they communicate, and how the overall solution is orchestrated.
This subsection synthesizes the ecosystem knowledge from Phase 1 with the strategic planning from this phase. The exam tests whether you can select the right platform for each agent based on requirements, and design the integration between them.
Platform Selection Framework:
Design Principles for Multi-Platform Solutions:
-
Start with the user experience. Where does the user interact with AI? If it's in Teams, start with Microsoft 365 Copilot extensibility. If it's in D365, start with embedded Copilot customization. If it's a standalone conversational experience, start with Copilot Studio.
-
Use the simplest platform that meets requirements. If D365 configuration provides the needed AI capability, don't build a custom agent. If Copilot Studio can orchestrate the logic, don't write code in Microsoft Foundry. Complexity is a cost.
-
Design integration points early. How will agents on different platforms share data and delegate tasks? Define the MCP servers, A2A connections, and connector configurations as part of the architecture, not as afterthoughts.
-
Plan for governance across platforms. A multi-platform solution needs unified governance. An agent built in Copilot Studio and a model deployed in Microsoft Foundry should both follow the same access control, monitoring, and responsible AI policies.
Common Multi-Platform Patterns:
| Pattern | Platforms | Use Case |
|---|---|---|
| Copilot Studio + D365 | Agent accesses D365 data via Dataverse connector | Customer service agent with CRM context |
| Copilot Studio + Foundry | Agent uses custom model hosted in Foundry | Specialized analysis (medical imaging, document classification) |
| D365 + Foundry | Custom model enriches D365 business process | Predictive demand forecasting in Supply Chain Management |
| M365 Copilot + Copilot Studio | Copilot Studio agent extends M365 Copilot | Company-specific knowledge agent available in Teams |
| All three | Supervisor in Copilot Studio orchestrates D365 agents and Foundry models | Enterprise AI hub spanning multiple business domains |
Reflection Question: A healthcare organization needs: (1) a patient-facing chatbot for appointment scheduling, (2) a clinical decision support tool using a custom medical model, and (3) AI-enhanced case management in D365 Customer Service. Design the multi-platform architecture.
