From Zero to Deployed Agent
Copilot Studio (formerly Power Virtual Agents) is Microsoft's low-code platform for building conversational AI agents. In 2026, it has evolved far beyond simple FAQ chatbots — it now supports generative orchestration, multi-agent coordination, and enterprise tool integration via MCP.
Core Building Blocks
| Component | Purpose | Example |
|---|
| Topics | Conversation flows triggered by user intent | "Track my order" topic with order lookup logic |
| Entities | Data types the agent recognises and extracts from user input | Order numbers, dates, product names |
| Knowledge Sources | Data the agent uses for generative answers (RAG) | SharePoint sites, Dataverse, websites, uploaded files |
| Plugins/Actions | External capabilities — Power Automate flows, connectors, custom APIs | Looking up CRM data, sending email confirmations |
| Generative AI | LLM-powered responses for open-ended questions | Answering HR policy questions from a knowledge base |
Publishing Channels
Once built, agents can be published to multiple channels simultaneously:
- Microsoft Teams: Embedded as a chat bot in Teams channels and personal chat (Note: "Classic" manual PVA Teams app deployment is deprecated; use Copilot extension deployment)
- Web Widget: Embeddable chat widget for any website
- Microsoft 365 Copilot: Appears as a plugin inside M365 Copilot
- Power Apps: Embedded within canvas and model-driven apps
- Real-Time Voice: Native deployment to Dynamics 365 Contact Center as conversational voice agents
- Facebook, Slack, Twilio: Third-party messaging integrations
- Custom API: Direct API integration for headless deployments
Getting Started
- Go to
copilotstudio.microsoft.com and create a new agent
- Define 2-3 key topics covering your most common user questions
- Add a knowledge source (e.g., a SharePoint document library) for generative answers
- Connect a Power Automate flow for any actions the agent needs to perform
- Test in the built-in chat panel, then publish to Teams
💡 Key Insight: Start with knowledge sources + generative answers before building custom topics. In many cases, pointing the agent at your documentation covers 70-80% of user questions without any manual topic authoring.