Skip to main content

Core Concepts

Copilot Studio assembles four building blocks into an agent: knowledge, tools, topics, and instructions. The platform's AI orchestrator uses these together to decide how to respond to any given user request.

Instructions​

Plain-language guidance that sets the agent's tone, defines what it should and shouldn't do, and establishes ground rules the orchestrator should follow — the closest thing to a system prompt in Copilot Studio.

Knowledge​

Knowledge sources ground the agent's answers in your organization's actual data — SharePoint sites, public websites, uploaded files, or enterprise data sources — instead of relying on the model's general knowledge. When a user asks a question, the orchestrator can pull an answer directly from a connected knowledge source.

Tools (Actions)​

Tools, also called actions, are what let an agent do something rather than just answer a question — calling an API, creating a record in a business system, or triggering a Power Automate flow. The orchestrator decides which tool to call, and with what inputs, based on the user's request.

Topics​

Topics are structured conversation paths — the older, hand-authored building block. A topic can define an exact flow for a specific, well-known scenario (like "reset my password"). Even as generative orchestration handles more of the conversation dynamically, topics remain useful for tightly-controlled, high-stakes conversation paths you don't want left entirely to the model's judgment.

Generative orchestration​

Generative orchestration is the planning layer that ties everything together: it interprets user intent, decides whether to answer from knowledge, trigger a topic, call a tool, or ask a clarifying question, and can chain multiple of these together to fully resolve a multi-part request. Topic triggers let you hook custom logic into specific points of this orchestration lifecycle when you need tighter control.

Workflows and agent-to-agent communication​

Workflows let you define structured, multi-step processes — and, as of 2026, can include other existing agents as steps. Agent-to-agent (A2A) communication extends this further, letting independently built agents exchange information and delegate tasks to one another, which is how Copilot Studio scales from a single conversational agent to a coordinated set of agents.

Next: Build your first agent walks through creating a simple agent end to end.