How to Pick the Best Pattern for Scalability When Building Agents in Copilot Studio

Welcome to this Cloud Wars Agent and Copilot Minute. In these discussions, I’ll be analyzing opportunities, impact, and outcomes possible with AI; today I present three patterns, or architectural approaches, you should know before building AI agents in Microsoft Copilot Studio. Choosing the right pattern at the start of a project is necessary to ensure the agent scales.

Highlights

00:23 — The first pattern is the orchestrator and specialist model, where a parent agent routes requests to specialized agents that handle specific areas well. Practical example: an employee help agent that routes questions about annual leave to the HR specialist agent and laptop issues to the IT specialist agent. This pattern is recommended for scopes spanning more than one business area.

01:35 — The second pattern is connected agents through Model Context Protocol, or MCP, where systems, data, and actions are exposed once as an MCP server. Example: Microsoft’s Dataverse MCP server, which allows multiple agents (sales, service, and operations) to access the same data without rebuilding queries. It is recommended whenever more than one agent needs access to the same data or actions.

02:30 — The third pattern is desktop or browser session use, where the agent operates a desktop or browser session the same way a person would. Example: processing supplier invoices in a legacy ERP with no API, where the agent opens the supplier portal, downloads the PDF, extracts line items, and keys them into the invoicing screen. This pattern is a last resort when there is no API or MCP server available and the work is repetitive.

More of my AI Insights:

• As Microsoft-OpenAI Deal Evolves, Customers Need to Focus on Outcomes Over Models
• How AI Changes the Role of Software From Providing Data to Suggesting Actions
• How to Align Microsoft Licensing With Work in the AI Era
• Microsoft Advances ‘Zero-Wait’ Enterprise for Retail
• Agent 365 Is Microsoft ‘HR for AI Agents’

For a 36-Hour Immersion into the FY27 Priorities that define Partner Success in the AI Era, join us at the AI Business Solutions Partner Executive Summit, running July 22-23, 2026, in Bellevue, Washington. Register today.