4.1.1. Built-in Copilot Capabilities vs. Agents
💡 First Principle: Built-in Copilot features are part of the M365 application experience — they're embedded in Word, Excel, Teams, Outlook, and other apps to assist with tasks in context. Agents are separate AI automation objects that can be scoped to a specific task, connected to specific data sources, and deployed to specific users or channels.
Built-in Microsoft Copilot capabilities (require a Microsoft Copilot add-on license):
| App | What Copilot Does |
|---|---|
| Word | Draft, rewrite, summarize documents |
| Excel | Analyze data, generate formulas, create charts, explain trends |
| PowerPoint | Create presentations from prompts or documents, summarize decks |
| Outlook | Draft emails, summarize threads, suggest replies |
| Teams | Answer questions in real time during a meeting; produce a post-meeting summary and action items; answer questions about chats and channel conversations. Copilot doesn't record the meeting itself: recording and transcription are Teams meeting settings, and Copilot can reference a meeting after it ends only when transcription or recording was on |
| Microsoft Copilot Chat (formerly Microsoft 365 Copilot Chat) | Cross-app chat assistant — web-grounded chat is included with eligible Microsoft 365 plans; grounding in your org data needs the Microsoft Copilot add-on license |
Agents are different in three key ways:
- They can be scoped — connected to specific SharePoint sites, specific data sources, or specific APIs
- They can be autonomous — some agents can take actions (send emails, update records) not just generate text
- They require separate governance — creation, testing, approval, and publishing are distinct steps
Declarative agents vs. custom engine agents. A declarative agent is a customized version of Copilot defined by configuration rather than code: its purpose and instructions (persona, tone, scope), conversation starters, the knowledge sources that ground it, and optionally actions or connectors. It runs on Copilot's own orchestrator and models, and can be built without writing code, for example in Copilot Studio. A custom engine agent instead uses its own orchestration and language model, chosen by the organization, for cases that need full control of the model, hosting or experience.
| Copilot (built-in) | Agent | |
|---|---|---|
| Scope | All M365 data user can access | Scoped to specific sources defined at creation |
| Created by | Microsoft | IT admins or users (via Copilot Studio) |
| Governed via | M365 admin center Copilot settings | Power Platform admin center + approval workflow |
| License | Microsoft Copilot license | Copilot license + agent usage (may be metered) |
⚠️ Exam Trap: Disabling a Copilot feature (like Copilot in Teams) doesn't disable agents that run in Teams. Features and agents are governed through different admin surfaces.
Reflection Question: A user wants to build an agent that answers HR policy questions by searching only the HR SharePoint site. Is this a Copilot feature or an agent? What tool do they use to build it?