Agents overview

Agents in Microsoft Copilot Studio are AI-powered assistants that understand your business context and take action on your behalf. You build agents on one of four harnesses: the GitHub Copilot harness for reasoning-heavy, multistep work, the Copilot chat harness for extending Microsoft Copilot Chat with your organization's knowledge, the Customer Conversation harness for customer-facing scenarios over voice and chat, or the standard harness for rule-based, predictable conversations. All harnesses let you combine generative AI, knowledge, and actions to create modern agents built for how your teams actually work.

You can deploy agents across multiple channels—from Microsoft Teams and web to enterprise applications. Agents can hand off to other agents, call workflows, and integrate with your business systems through connectors, REST APIs, and MCP servers.

Note

This article applies to multiple harnesses. The harness you use affects the billing, features, and capabilities of what you build.

Benefits of agents

Agents offer several benefits.

  • Reasoning-heavy automation: GitHub Copilot harness agents think through complex, multistep business processes, break them into steps, and adapt when circumstances change. They work like a capable teammate, not just a script.
  • Predictable responses: Standard harness agents follow the topics and rules you define, providing consistent answers to well-understood questions with no surprises.
  • Action-ready: Agents can automate tasks by calling workflows, connectors, REST APIs, MCP servers, and other agents. They connect to your business systems to take real action.
  • File-ready: GitHub Copilot harness agents can create, edit, and reason over Word, Excel, PowerPoint, and PDF files natively, making them ideal for document-intensive processes.
  • Conversational context: Both the GitHub Copilot and standard harnesses support natural conversations with memory and context awareness so agents improve over time.
  • Microsoft Copilot integration: Copilot chat harness agents connect your enterprise knowledge to Microsoft Copilot Chat so employees get grounded answers without leaving their everyday experience.
  • Contact center automation: Customer Conversation harness agents combine a primary agent with specialized sub-agents to handle customer questions over voice and chat, escalating to a human representative when the agent can't resolve the issue.

Four harnesses, four approaches

All agents in Copilot Studio run on one of four harnesses, each designed for different scenarios.

GitHub Copilot harness

The GitHub Copilot harness powers agents built for reasoning-heavy, multistep business processes. These agents break down complex goals into steps, reason through problems, recover from errors, and adapt to changing requests. They're ideal for:

  • Complex business processes: Accounts payable, contract review, hiring workflows, or other multistep processes that require judgment and adaptation.
  • File-intensive work: Agents that create, edit, and reason over documents.
  • Multitool orchestration: Tasks that require calling several tools or agents in sequence and deciding what to do based on results.
  • Agentic reasoning: Work that benefits from the agent thinking through a problem rather than following a fixed script.

GitHub Copilot harness agents have native support for skills and memory, and run in a secure sandbox. They use Copilot Credits for billing. Learn more in Overview of billing for agents powered by the GitHub Copilot harness.

Copilot chat harness

The Copilot chat harness powers agents that extend Microsoft Copilot Chat with your organization's knowledge. These agents connect enterprise content to the Microsoft Copilot Chat experience so employees get answers grounded in your data without leaving their everyday tools. They're ideal for:

  • Knowledge-first scenarios: Employee onboarding agents, policy FAQ agents, or any scenario where the priority is connecting people to information.
  • Microsoft Copilot extension: Adding domain-specific or company-specific knowledge to the Microsoft Copilot Chat experience.
  • Internal publishing: Reaching employees directly through Microsoft Copilot without building a separate channel.

Copilot chat harness agents run on current chat models and publish to internal teams. Billing is consumption-based or included in a Microsoft Copilot user subscription license. Learn more in Licensing for agents powered by the standard harness.

Customer Conversation harness

The Customer Conversation harness powers agents that answer questions and resolve issues for customers over voice and chat. A primary agent works with specialized sub-agents in a multi-agent architecture to provide predictable responses for defined topics while still supporting fluid, natural-language conversation. They're ideal for:

  • Contact center scenarios: Customer support, help desks, and other scenarios where a customer needs to resolve an issue over voice or chat.
  • Multi-agent routing: Scenarios where different topics or issues are best handled by specialized sub-agents, each with its own escalation path.
  • Human escalation: Conversations that may need to hand off to a human representative when the agent can't resolve the issue.

Customer Conversation harness agents can call tools across connectors, knowledge, MCP, and connected agents, and support skills and memory. They use Copilot Credits for billing. Learn more in Agents powered by the Customer Conversation harness overview.

Standard harness

The standard harness powers rule-based agents built for structured, repeatable conversations. These agents follow the topics you define and respond predictably to well-understood questions. They're ideal for:

  • Help-desk and support: Internal help agents that answer common questions and route simple requests.
  • Information lookup: Agents that provide consistent answers based on your enterprise knowledge.
  • Structured conversations: Scenarios where you want the agent to guide the conversation predictably.

Standard harness agents work with existing prompt libraries and enterprise knowledge. They use the Copilot Studio capacity model for billing. Learn more in Licensing for agents powered by the standard harness.

  • Ultima actualizare pe