platform-architecture

Microsoft Frontier Agent Platform Architecture

This skill provides comprehensive knowledge of the Microsoft Frontier Agent Platform — the full stack that powers AI agents across the Microsoft ecosystem. The platform has 10 interconnected layers, and understanding how they fit together is essential for designing effective agent solutions.

Platform Layers Overview

The Frontier Agent Platform is organized into these interconnected layers. Consult references/platform-layers.md for detailed descriptions of each component.

Layer Map (Top to Bottom)

1. The UI for AI — Microsoft 365 Copilot as the front door for all agents
2. Agent Store — 3rd party agents, Microsoft agents, and custom agents
3. Agent Hosting / Orchestration — M365 Copilot, Copilot Studio, Foundry, Windows 365
4. Intelligence — Work IQ, Fabric IQ, Foundry IQ (personalization, memory, orchestration)
5. Open Protocols — OpenAPI, MCP, A2A for interoperability
6. MCP Servers — Outlook, Teams, Search, Fabric, D365 (Microsoft's first-party MCP servers)
7. Connectors — D365, Salesforce, SAP, Workday, ServiceNow, Dataverse
8. Enterprise Data Platform — Azure Databricks, Microsoft Fabric, Dataverse
9. Enterprise Systems of Record — D365 CRM, D365 F&O, Salesforce, SAP, Workday, ServiceNow, SQL, Storage
10. Agent Control System — Agent 365, Copilot Analytics, Operations (side panel)

Agent Development Tooling (Left Panel)

The development side spans a spectrum from no-code to full pro-code:

• Power Platform: Copilot Studio Lite, Copilot Studio Full, Power Automate, Power Apps
• Azure AI: Foundry Agent Service, Logic Apps
• Agents Toolkit: GitHub Copilot (Agent Mode), M365 Agents SDK, Agent 365 SDK, Agent Framework

Architecture Design Workflow

When helping design a solution architecture:

Step 1: Gather Requirements

Identify the scenario, user personas, data sources, integration points, compliance needs, and scale expectations. Use references/requirements-template.md for a structured intake.

Step 2: Map to Platform Layers

For each requirement, identify which platform layer(s) are involved. A typical agent solution touches 4-6 layers. Consult references/platform-layers.md for component capabilities.

Step 3: Select Components

Choose specific components within each layer. Common patterns include:

Pattern	Components	Best For
Simple Q&A Agent	Copilot UI → Agent Builder → SharePoint	Document retrieval, FAQ
Business Process Agent	Copilot UI → Copilot Studio → Power Automate → D365	Workflow automation
Pro-Code Agent	Copilot UI → Agents SDK → Foundry → Azure AI	Custom orchestration
Multi-Agent System	Copilot UI → Custom Engine → A2A → Multiple Agents	Complex workflows
Data Intelligence Agent	Copilot UI → Fabric IQ → Microsoft Fabric → Dataverse	Analytics, BI

Step 4: Design Integration Points

Map how components communicate. Key integration protocols:

• MCP (Model Context Protocol) — Agent-to-tool communication
• A2A (Agent-to-Agent) — Multi-agent orchestration
• OpenAPI — REST API integration
• Microsoft Graph — M365 data access
• Dataverse — Structured business data

Step 5: Address Cross-Cutting Concerns

For every architecture, address:

• Identity & Access: Entra ID, least-privilege, conditional access
• Security: Purview for data governance, Defender for threat protection
• Compliance: Guardrails, policies, audit trails
• Monitoring: Copilot Analytics, operational dashboards
• Quota & Rate Limiting: TPM, PTU, rate-limiting considerations

Step 6: Create Architecture Artifacts

Generate visual architecture diagrams using Mermaid or draw.io format. Use references/diagram-templates.md for reusable architecture diagram patterns.

Using Microsoft Learn MCP for Latest Information

Always verify architecture details against current documentation. Use microsoft_docs_search and microsoft_docs_fetch to:

• Confirm component names and capabilities (the ecosystem evolves rapidly)
• Check GA/preview/deprecated status of features
• Verify integration patterns and supported protocols
• Retrieve current pricing and licensing models

Reference Files

• references/platform-layers.md — Detailed breakdown of all 10 platform layers with every component
• references/requirements-template.md — Structured requirements gathering template
• references/diagram-templates.md — Reusable Mermaid architecture diagram templates
• references/integration-patterns.md — Common integration patterns between layers