Understand generative AI fundamentals
Copilot Context, Experiences, and Agents
CoreChoose the Copilot surface and context that fit a business task while respecting organizational data and permissions.
Aligned to the AB-730 skills measured as of July 22, 2026; product behavior verified August 23, 2026.
Why this matters
AB-730 expects a business user to choose between broad chat, an app-native experience, an available agent, and a tailored agent before prompting. That choice determines which work, web, app, and conversation context can shape the response.
Must Know
- Organizational Copilot prompts and responses receive enterprise data protection. Existing identity, permission, sensitivity, retention, audit, and administrative controls still apply; Copilot does not grant new access.
- Prompts, responses, and Microsoft Graph data accessed by organizational Copilot are not used to train foundation models. This privacy commitment does not mean every source is available to every user.
- Context may come from the current app or document, the conversation, authorized work data, and the web. Deliberately referencing a source narrows grounding to material chosen for the task.
- Use chat for broad or one-time conversational help. Use an agent when specialized knowledge, instructions, or repeatable behavior is the deciding requirement.
- Check Agent Store for a suitable available agent before creating one. Build a tailored agent only when the available choices cannot meet an important knowledge or behavior requirement.
- Choose the app-native Copilot surface when the task depends on that app’s content or workflow: Word for documents, Excel for structured analysis, PowerPoint for presentations, Outlook for mail, and Teams for meetings and collaboration.
Compare and Distinguish
- Chat vs agent: chat is flexible conversation; an agent is configured for a specialized purpose with selected knowledge and behavior.
- Agent Store vs custom agent: discover and reuse a suitable existing agent; create only to satisfy a real tailored requirement.
- Implicit context vs explicit reference: the open app, prior turns, and available work/web data can supply context; a reference deliberately identifies the file, meeting, email, person, or other resource that should ground the task.
- App-specific Copilot vs general chat: use the app surface when native content and actions matter; use general chat for broader, cross-work assistance.
- Researcher vs Analyst: Researcher supports complex multi-source research and source-cited reports; Analyst supports data-heavy analysis such as trends, statistics, outliers, and visual insights.
Scenario examples
- Scenario: A manager wants one quick rewrite of an email. Think: Use chat or the Outlook-native experience, not a new agent.
- Scenario: HR repeatedly answers onboarding questions from one approved handbook in a consistent style. Think: Evaluate Agent Store, then create a tailored agent if no available agent meets the knowledge and instruction requirements.
- Scenario: A finance analyst needs trends and outliers from a structured workbook. Think: Use Excel or Analyst rather than a Word drafting workflow.
- Scenario: A proposal must reflect the open Word document. Think: Work in Word, name the desired audience and output, and reference the approved source explicitly when source control matters.
Exam traps
- Copilot cannot retrieve a file merely because the prompt asks clearly; the user still needs legitimate access.
- More context is not automatically better. Irrelevant, stale, or sensitive context can reduce suitability or create risk.
- An agent is not simply a renamed chat, and a long saved prompt does not provide the same reusable knowledge-and-instruction configuration.
- Product experiences are related but not interchangeable; meeting evidence belongs in the Teams meeting context, while workbook analysis belongs in Excel.
- Current Microsoft sources use both Microsoft Copilot and Microsoft 365 Copilot during a naming transition. Branding trivia is not the tested decision.
Key takeaways
- Task evidence determines context; task workflow determines the best Copilot surface.
- Chat handles broad, ad hoc work; agents handle specialized, repeatable work.
- Reuse a suitable agent before building, and never confuse Copilot access with permission expansion.
How it works
- Copilot combines the user’s request with available context, applies the user’s access boundary and organizational controls, and generates a response that still requires review.
- A configured agent adds a declared purpose, knowledge, instructions, and supported capabilities to make a class of tasks more consistent than a blank conversation.
Objects and administrative surfaces
- Microsoft Copilot Chat — broad work conversation, prompting, web/work grounding where available, and access to agents.
- Word, Excel, PowerPoint, Outlook, and Teams — app-native experiences that use the current item or workflow as relevant context.
- Agent Store — discovery and use of available agents; Agent Builder — lightweight creation and customization for scenario-specific needs.
- SharePoint, OneDrive, email, chats, meetings, and other Microsoft 365 resources — possible work context only when the user and the Copilot experience are allowed to access and process them.
When to use it
- Use chat for brainstorming, rewriting, quick questions, and exploratory follow-ups.
- Use an app-native surface when the open document, workbook, presentation, message, or meeting is central.
- Use an existing agent for a matching specialized capability; create one for an unmet, repeatable, bounded need.
Security and governance implications
- Stay in the approved organizational experience for work information and inspect an agent’s terms when relevant.
- Reference only necessary, authorized sources and do not assume a recipient of an agent or output gains access to its underlying data.
How to reason about this
- If a response lacks expected work facts, check the chosen surface, source reference, source freshness, user permission, and protection restriction before broadening the prompt.
- If an agent responds generically, determine whether the gap is knowledge, instructions, capability, or simply an unsuitable agent choice.
More detail
- The organizational Copilot service uses the signed-in user’s security context. Microsoft Graph can supply work context, but only through applicable access and product controls.
- Web grounding can add current public information, while work grounding can add authorized organizational information. Verify which source supports each decisive claim.
- Agent Builder supports straightforward scenario-specific agents with dedicated knowledge sources and testing. Advanced integrations and developer mechanics are outside AB-730 business-user scope.
- Researcher and Analyst are named in the audience profile as agents a candidate should recognize; they do not add objectives beyond the published 30 skills.
Ready for the quiz?
- When does app context matter more than general chat?
- What requirement justifies creating an agent instead of using chat or Agent Store?
- Why can Copilot fail to use a file that the user named?
- Which clue points to Researcher rather than Analyst?
Related objectives
- D1.1.a — Understand how Copilot works to keep your organization’s information private and secure
- D1.1.b — Understand how the context, like your work files, web data, or the app you’re using, can affect Copilot responses
- D1.1.c — Understand the difference between a chat experience and an agent experience
- D1.1.d — Understand the use case for creating your own agent
- D1.1.e — Understand the differences in features and capabilities of the Copilot experience in various Microsoft 365 apps