Microsoft shares API-powered agent insights and implementation guide

Microsoft has shared its internal learnings on deploying API-based agents with Microsoft 365 Copilot, highlighting their potential to streamline workflows and enhance productivity. The company's IT organization, Microsoft Digital, has been experimenting with these agents to access diverse data sources beyond Microsoft Graph, eliminating the need to build agents from scratch.

The Need for API-Based Agents

Microsoft 365 Copilot serves as the enterprise UX, with agents acting as background applications that save time and improve efficiency. While Copilot's out-of-the-box access to Microsoft Graph suffices for many users, scenarios requiring data from other sources necessitate an API layer. This approach allows organizations to extend Copilot's reasoning capabilities without fine-tuning models.

"Cost is one of the most critical dimensions in how we design, deploy, and scale our solutions," says Faisal Nasir, Principal Architect at Microsoft Digital. "Declarative API-driven agents in Microsoft 365 Copilot offer a path to unify agentic experiences while leveraging shared AI compute and infrastructure."

Learn more about Microsoft's agent deployment strategy.

Case Study: Azure FinOps Budget Agent

The Azure FinOps Budget Agent exemplifies the power of API-based agents. Designed to help teams track Azure spending, it replaces traditional dashboards with a natural language interface. The agent pulls data from:

• SQL Server for budget and forecast data
• Azure Spend for actual spending
• Projected spending services

Users can query the agent for specific insights, such as:

• "Get me the monthly breakdown of service Azure Optimization Assessment analytics."
• "Which service deviates most from its budget forecasts?"

The agent has helped Microsoft save 10–12% of its Azure cost footprint and significantly reducing analysis time for users.

Faris Mango, Principal Software Engineering Manager, Microsoft Digital

Five Key Strategies for Building API-Based Agents

1. Start with user intent, not the API