Create 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.
Choose an authoring tool
Microsoft provides multiple tools for building agents. The authoring tool determines how you configure and manage an agent. It isn't the same as the harness that runs the agent.
| Authoring tool | Primary audience | Use it for |
|---|---|---|
| Agent Builder in Microsoft 365 Copilot | Information workers | Quick, knowledge-focused declarative agents for individuals and teams |
| Microsoft Copilot Studio | Makers and low-code developers | Agents and workflows that use enterprise knowledge, tools, governance, and multiple publishing channels |
| Microsoft Foundry | Professional developers | Code-first agents that require custom models, logic, infrastructure, or application integration |
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.
Choose a Copilot Studio harness
Every agent or workflow that you build in Copilot Studio runs on a harness. The selected model provides reasoning and content generation. The harness is the runtime between the model and the components of your agent. It decides when to call the model, which instructions and context to send, how to interpret the result, and which tools to call.
Copilot Studio supports three harnesses:
| Consideration | GitHub Copilot harness | Standard harness | Copilot chat harness |
|---|---|---|---|
| Best for | Reasoning-heavy, multistep business processes | Rule-based agents and structured, repeatable conversations | Extending Microsoft 365 Copilot Chat with organizational knowledge |
| How it works | Breaks a goal into steps, calls tools and agents, and adapts when conditions change | Uses topics, rules, knowledge, and configured orchestration | Grounds Microsoft 365 Copilot Chat in connected enterprise content |
| Distinct capabilities | Native file creation and editing, skills, memory, connected agents, and a secure task sandbox | Authored topics, deterministic paths, generative or classic orchestration, and agent flows | Knowledge-first agents that run in Microsoft 365 Copilot Chat |
| Publishing scope | Internal teams or external customers | Internal teams or external customers | Internal teams |
| Billing model | Copilot Credits | Copilot Studio licensing and capacity | Consumption-based or included in a Microsoft 365 Copilot user subscription license |
Note
The exercises in this module use the standard harness.
Choose the GitHub Copilot harness when an agent must reason across a long task, use several tools, work with files, or recover from a failed step. Choose the standard harness when you need predictable topic-driven behavior or structured conversations. Choose the Copilot chat harness when employees need grounded answers from organizational knowledge in Microsoft 365 Copilot Chat.
Important
You choose the harness when you create an agent. You can't transfer an existing agent between the GitHub Copilot and standard harnesses.
For current capabilities and billing considerations, see Choose a harness.
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.
When to use Copilot Studio
Use Copilot Studio when your agent needs one or more of these capabilities:
- Workflows, custom actions, or integration with business systems.
- Enterprise knowledge sources and grounded answers.
- Publishing to Microsoft Teams, Microsoft 365 Copilot, websites, or other supported channels.
- Power Platform governance, data policies, security roles, and application lifecycle management.
- Connected agents or multistep 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.
Note
This module doesn't discuss Microsoft Foundry agents further.
Next, you examine how Agent Builder creates declarative agents for Microsoft 365 Copilot Chat.