Skip to main content
Creating an agent in Copilot Studio. This guide shows how to create a custom agent using Microsoft Copilot Studio (the full authoring experience) or the quicker Copilot Chat flow. You can start in Copilot Chat — for example, create a “US Holidays in 2026” agent directly from chat — and then move into Copilot Studio for full configuration, governance, and publishing. When you click Create an agent in Copilot Chat, Copilot Studio opens under the hood to build the agent. All authoring and configuration occur in Copilot Studio. There are two primary approaches for creating a new agent in Copilot Studio:
  • Describe experience — Describe in natural language what you want the agent to do and Copilot Studio converts that into structured instructions and a baseline configuration. Use this for rapid prototyping and simple use cases (for example: “Create an agent that answers HR questions and helps employees find company policies”).
  • Configure experience — Use this mode to manually adjust name, description, instructions, knowledge sources, model selection, triggers, behavior rules, and other advanced settings. This experience is better for governance, auditing, and repeatable, predictable behavior.
A tutorial slide titled "Creating an Agent in Copilot Chat" showing three numbered steps (Copilot Studio, "Describe" Tab, "Configure" Tab) on the left and a screenshot of the Copilot Studio interface with agent templates (Career Coach, Customer Insights Assistant, Idea Coach) on the right. The page appears to guide users through building and configuring chat agents.
Information entered via the Describe experience is converted into the structured instructions that you can view and refine in the Configure tab. This lets you start with a simple conversation and then switch to a detailed, auditable configuration for production use.
Start with the Describe path for fast prototyping; then switch to Configure to review, refine, and apply governance controls before publishing.
For enterprise deployments, the Configure experience is typically preferred because it enables review, auditing, troubleshooting, and governance. Configure exposes the logic that drives the agent and supports consistent management practices across the organization. To open Copilot Studio, browse to copilotstudio.microsoft.com. If you’re signing in for the first time you may see a licensing prompt; you can sign up for a trial to evaluate Copilot Studio for learning and testing.
If you don’t have the required license, Copilot Studio will prompt you to sign up for a trial. Trials are intended for evaluation—confirm your organization’s licensing before deploying production agents.
The Copilot Studio start page gives you options to create an Agent or a Workflow and includes starter cards and recently-used items. You can create an agent from scratch or begin with a Describe prompt.
A screenshot of Microsoft Copilot Studio's web interface asking "What would you like to build?" with options to create an Agent or Workflow and a large input field to describe the agent. Below are starter cards for Agent, Computer-using agent, and Create workflow, plus a left sidebar menu and a "Recent agents" list.
Key capabilities of the Copilot Studio authoring experience include:
A screenshot of Microsoft Copilot Studio showing an "HR Agent" setup page with a model-selection dropdown listing GPT and Claude models. The right side shows a test chat pane for interacting with the agent.
Adding knowledge sources is straightforward. You can upload documents or connect to enterprise stores so the agent can reference relevant content when responding. Common knowledge sources to connect:
A screenshot of Microsoft Copilot Studio with an "Add knowledge" dialog open for an HR Agent. The dialog shows file upload and sync options plus connectors like OneDrive, SharePoint, Azure AI Search, Dataverse, Dynamics 365, Salesforce, ServiceNow, and Azure SQL.
Advanced options let you extend the agent’s reach and behavior:
  • Web search — allow live web lookups for up-to-date facts.
  • Work IQ integration — surface content from Outlook, SharePoint, Teams, and other Office 365 sources.
  • Triggers — create autonomous agents using event-based triggers (SharePoint, OneDrive, Teams, Outlook, Forms, Dataverse, Planner, etc.) or scheduled recurrences. This enables agents to run workflows or take action without a manual chat prompt.
A screenshot of Microsoft Copilot Studio showing an "Add trigger" dialog with a list of automation triggers (Recurrence, SharePoint, OneDrive, Teams, Outlook, Microsoft Forms, Dataverse, Planner, etc.) for an HR Agent. The underlying HR Agent workspace with Publish, Settings and Test buttons is faintly visible in the background.
Copilot Studio also supports multi-agent scenarios. You can chain agents, hand off tasks (for example, an HR agent transferring a finance question to a Finance agent), and define topics and suggested prompts to guide users toward the best interactions. Use the integrated test pane to validate responses, edge cases, and how knowledge sources are used. When the agent meets your requirements, publish it to your organization. Published agents can be managed, monitored, and updated centrally — making it easier to apply governance, measure usage, and iterate. Further reading and resources: Further topics you may want to explore next: managing published agents, monitoring usage and telemetry, applying governance policies, and designing multi-agent workflows.

Watch Video