Creating agents

Before you start creating an agent, it's important to consider what it's used for. For example, you might use it to manage account inquiries, or you could use it for self-service support cases such as knowledge base access. Knowing how you plan to use the agent helps you define and plot out conversation paths and determine how many topics the agent handles.

Other functions that you could consider include using it to look up basic account details, perform more advanced account operations, or implement some type of action. The more scenarios that you initially consider, the easier it is to determine the details that your agent needs for you to facilitate it.

Types of agents

Microsoft now offers three major agent‑building platforms, each designed for different audiences, capabilities, and levels of complexity. These platforms create different types of agents suited to distinct business scenarios.

Agent architecture and components

Most agents include a client interface, foundational infrastructure, an orchestrator, a language model, knowledge, and tool calling capabilities. This standard architecture provides the framework for implementing agents across different Microsoft platforms.

The differences between the platforms are essentially how much control you have over each of the components and the tools available to configure the individual elements.

Microsoft 365 Copilot (Agent Builder)

Agent Builder creates simple, fast, low‑maintenance agents inside Microsoft 365 Copilot. These agents:

• Answer questions using your team’s documents, emails, SharePoint, or OneDrive content.

• Provide consistent, approved answers for small teams or individual use.

• Require no code and are designed for everyday information workers.

  Note

  Agents created in Microsoft 365 Copilot are called declarative agents.

Microsoft 365 Copilot agents are best for:

• Personal or team‑level assistants
• Knowledge lookup
• FAQ‑style agents
• Internal use only

Microsoft 365 Copilot agents are focused on using knowledge from the Microsoft Graph, i.e., your daily work such as emails, Teams meetings, Teams chats, and files from SharePoint and OneDrive.

Copilot Studio (Low‑code / No‑code platform)

Copilot Studio produces agents that support structured tasks and enterprise integration. These agents can:

• Run multi‑step workflows.
• Use topics for guided conversations.
• Integrate with business systems via connectors, APIs, and automation.
• Use knowledge, tools, and generative orchestration.
• Be published broadly e.g., Microsoft Teams, web, messaging apps, and external audiences.

Copilot Studio agents are best for:

• Department or organization‑wide agents.
• Customer‑facing agents.
• Agents requiring compliance, governance, analytics.
• Process agents that perform actions, gather data, or automate tasks.

Why you should build agents with Copilot Studio

Microsoft Copilot Studio is purpose‑built for organizations that want secure, scalable, enterprise‑ready AI agents—far beyond what lightweight tools like Agent Builder can provide. The core reasons fall into five categories: capability, control, integration, governance, and business impact.

• Build more powerful, flexible agents - Copilot Studio supports complex, multi‑step workflows, custom actions, integration with business systems, and autonomous capabilities—not just simple Q&A.
• Designed for enterprise‑wide deployment - Copilot Studio enables publishing agents to departments, the whole organization, or external customers, across multiple channels such as Teams, web, apps, and Microsoft 365 Copilot.
• Deep integration with business systems and data - Copilot Studio provides 1,400+ connectors, Model Context Protocol (MCP), APIs, workflows, Dataverse, SharePoint, and more. Agents can incorporate enterprise knowledge sources and use RAG or workflows to return grounded answers.
• Strong governance, security, and lifecycle management - Copilot Studio provides the enterprise‑grade controls required by IT such as Governance, deployment controls, and application lifecycle management not available in lightweight agent builders.
• Built to support agentic business transformation - Copilot Studio enables organizations to automate processes, redesign workflows, and build multi-agent solutions. Copilot Studio provides speed-to-value without sacrificing control or quality.

You should build agents with Copilot Studio when you need:

• Enterprise‑grade, scalable AI agents.
• Advanced workflows, knowledge integration, and system actions.
• Deployment across Teams, web, and Microsoft 365.
• Strong governance, security, and lifecycle management.
• A platform that supports organization‑wide AI transformation.

Copilot Studio is the right choice when you need more than a simple Q&A agent—you need a reliable, secure, powerful agent that can automate real business processes.

Microsoft Foundry

Microsoft Foundry creates AI‑engineered, code‑first agents with full control over:

• AI model selection and fine‑tuning
• Data orchestration and pipelines
• Tool invocation
• Complex multi‑agent or autonomous workflows
• Security, compliance, and custom infrastructure
• Integration with cloud systems and applications

Microsoft Foundry agents are best for:

• Large enterprises needing full model, data, and system control.
• Highly specialized agents requiring custom logic or advanced autonomy.
• Multi‑agent systems, deep integration, or enterprise‑scale workloads.

We'll now look at how to create declarative agents in Microsoft 365 Copilot.