Note
Access to this page requires authorization. You can try signing in or changing directories.
Access to this page requires authorization. You can try changing directories.
[This article is prerelease documentation and is subject to change.]
Create an agent to design multi-agent orchestrations for customer-facing scenarios, such as contact centers and help desks. An agent built with this harness starts with a primary agent that greets customers, classifies intent, and routes conversations to specialist sub-agents that each handle a focused task. Every agent you create this way supports both Speech and Chat conversations by default, so you don't need to choose a channel modality up front.
Note
The harness used to build this agent uses usage-based billing. Using, building, testing, and evaluating agents might consume Copilot Credits. Learn more in Overview of usage-based billing.
Create from Copilot Studio or the Microsoft Contact Center Intelligence portal
You can create a Customer Conversation harness agent from either Copilot Studio or the Microsoft Contact Center Intelligence portal. Both surfaces use the same underlying builder, so the authoring experience (instructions, skills, tools, knowledge, events, and sub-agents) is identical no matter where you start.
- Copilot Studio: Use the steps in this article. Choose this entry point if you're already working in Copilot Studio, or if you don't yet need to connect the agent to a live contact center.
- Microsoft Contact Center Intelligence portal: The Microsoft Contact Center Intelligence portal is where you connect an agent to a contact center (such as Dynamics 365 Contact Center, Genesys, Cisco, or Avaya), configure biometric authentication, and manage quality, insights, and governance for the agent. Deploying an agent to a contact center can only be done from the portal, by an Omnichannel Admin. For steps to create an agent from the Microsoft Contact Center Intelligence portal, see Create a Customer Assist Agent.
This article covers only the Copilot Studio entry point. It doesn't include steps for creating an agent from the Microsoft Contact Center Intelligence portal.
To create an agent from Copilot Studio:
Sign in to Copilot Studio.
Start from one of the following entry points:
- On the Home page, under Other ways to build, select the Customer Conversation tile.
- On the Agents page, select New agent, then select Agent from the list, shown with a Customer Conversation badge.
Describe the scenario you want your agent to handle in the text box. Copilot Studio uses your description to generate a starting primary agent, including a suggested name, description, and instructions, along with suggested triggers, channels, knowledge, and tools. Or, under Start from scratch, select Create agent to start from a blank primary agent instead.
Note
You can also describe your agent through a multi-turn conversation with Copilot Studio, refining the scenario over several messages before it generates your starting agent.
Review and adjust the generated fields for your primary agent:
- In the Name field, confirm or update the name for your primary agent.
- In the Description field, confirm or update what the agent handles, so the orchestrator knows when to route conversations to it. This field has a 1,024-character limit.
- In the Instructions section, confirm or update how the agent should behave, including its role, tone, and any rules or guidelines it must follow. General instructions have a hard limit of 20,000 characters. Because every agent supports both Speech and Chat conversations, you can add modality-specific guidance in the Speech and Chat tabs of the Instructions section. Each of these tabs also has a hard limit of 20,000 characters. Learn more in Configure agent details and instructions (Customer Conversation).
On the Build tab, select the primary agent's card to open its configuration panel. If you've added any sub-agents, select a sub-agent's card instead to configure that sub-agent. Continue building the selected agent:
- Add skills. Skills define what the agent can do. Learn more in Skills overview and Create a skill (Customer Conversation).
- Add tools. Tools configure how your agent connects to services, including prebuilt tools like Escalate. Voice capabilities include Biometric Authentication for caller verification. Learn more in Prebuilt contact center tools (Customer Conversation), Add prebuilt contact center tools (Customer Conversation), and Tools overview.
- Add knowledge sources to ground the agent's responses. Learn more in Add knowledge to an agent (Customer Conversation).
- Configure events, such as Conversation Start, End of Conversation, and On Error, to control how the agent responds at key moments in a conversation. Learn more in Events overview (Customer Conversation) and Configure events (Customer Conversation).
Add specialist agents to your orchestration:
- Select Add a sub-agent to create a new specialist agent that handles a focused intent, such as billing or technical support. A sub-agent shares configuration, deployment, and application lifecycle management (ALM) with its primary agent. Learn more in Sub-agents overview (Customer Conversation) and Add a sub-agent (Customer Conversation).
- Or, select Connect existing agent to add an independent agent, such as an agent hosted in Azure AI Foundry, to the orchestration. A connected agent keeps its own configuration and ALM, so you can reuse it across multiple orchestrations. You can remove a connected agent to preserve the connection for later, or delete it to permanently remove it. Learn more in Connected agents overview.
After building your agents, you can also test your orchestration. Learn more in Preview and test an agent (Customer Conversation) and Evaluate an agent (Customer Conversation).
When your agent is ready, publish it to make it available for use and monitor its performance. Learn more in Deploy overview (Customer Conversation) and Monitor your agent (Customer Conversation).