Perform basic administrative tasks for Copilot and agents
Copilot, Researcher, Analyst, and Custom Agents
CoreChoose standard Copilot, first-party agents, SharePoint agents, or custom agents according to task depth, data, repeatability, and action requirements.
Aligned to the current AB-900 guide, verified July 22, 2026.
Why this matters
AB-900 expects you to choose the simplest Copilot or agent experience that matches the task, source, audience, and governance need.
Must Know
- Standard Microsoft 365 Copilot supports broad everyday drafting, summarizing, meeting, and question-answering tasks.
- Researcher is for complex, multi-step research that produces a structured, source-cited report from permitted work and web sources.
- Analyst is for data reasoning, calculations, trends, anomalies, and analytical outputs.
- A saved prompt is reusable instruction text; an agent is a maintained experience with purpose, owner, audience, knowledge, instructions, access, and lifecycle.
- A SharePoint agent provides focused Q&A over selected SharePoint knowledge while honoring each user’s source permissions.
- Use a custom agent when the organization needs a repeatable specialized purpose; add connectors or actions only when the task requires them.
- Prebuilt and custom agents still depend on identity, source permission, security, and compliance controls.
Compare and Distinguish
- Copilot vs agent: broad assistance for varied tasks vs persistent specialization for a defined purpose.
- Researcher vs Analyst: multi-source, source-cited research report vs data reasoning, calculations, trends, and analytical outputs.
- Prebuilt vs custom agent: Microsoft-provided specialist vs organization-defined purpose, knowledge, and governance.
- SharePoint agent vs advanced custom agent: focused SharePoint knowledge experience vs potentially broader data, actions, integrations, and workflow.
- Saved prompt vs agent: reusable prompt text executed by a user vs a maintained experience with ownership, knowledge, instructions, access, and lifecycle.
Scenario examples
- Scenario: A user needs a short summary of today’s meeting. Think: Standard Copilot is sufficient.
- Scenario: Leadership needs a structured, cited market briefing across work and web sources. Think: Use Researcher.
- Scenario: Finance needs trends and anomalies across several datasets. Think: Use Analyst.
- Scenario: A department wants a repeatable assistant grounded in approved procedures. Think: Create an agent instead of asking everyone to maintain a long prompt.
- Scenario: A site needs a Q&A experience limited to selected SharePoint knowledge. Think: A SharePoint agent is the focused choice.
Exam traps
- Researcher is not the best choice for a quick email summary, and Analyst is not primarily a web research writer.
- A custom agent is unnecessary for every one-off prompt.
- A saved prompt is not an agent and has no independent knowledge/action lifecycle.
- Agent specialization does not bypass Microsoft 365 or connector permissions.
Key takeaways
- Quick/general → Copilot; deep/cited → Researcher; data reasoning → Analyst.
- Repeatable purpose → agent; SharePoint-focused knowledge → SharePoint agent.
- Choose by task, sources, actions, audience, and governance—not by the newest name.
How it works
- Standard Copilot interprets each prompt with permitted context. An agent adds persistent instructions, selected knowledge, and optional capabilities so repeated interactions follow a more consistent purpose.
- Researcher deliberately spends more time retrieving and reasoning across sources. Analyst uses analysis-oriented capabilities to work through data. Both remain subject to access and administrative availability.
- SharePoint and custom agents retrieve only sources the current user may access, and any action/connector also operates under its configured authentication and authorization model.
Objects and administrative surfaces
- Microsoft 365 Copilot Chat and Microsoft 365 apps — standard Copilot and first-party agent experiences.
- SharePoint sites/libraries and .agent files — SharePoint agents and their access.
- Agent Builder — lightweight in-context creation; Microsoft Copilot Studio — richer low-code agents and workflows.
- Agent Registry/Agent 365 registry and Copilot settings — Microsoft 365 admin center; Power Platform inventory — environment-aware org-built agents.
When to use it
- Use standard Copilot for fast, varied individual productivity tasks; use Researcher when the outcome is a deeper cited report.
- Use Analyst when the main work is quantitative/data analysis and choose a custom agent for repeatable approved knowledge or workflow.
- Use a SharePoint agent when selected SharePoint content is the intended knowledge boundary.
Security and governance implications
- Review the agent’s owner, purpose, audience, knowledge, instructions, actions, permissions, and data path before broad availability.
- Use existing source permissions and least-privileged connector/action access; do not make data public to simplify agent design.
- Treat first-party agent output as generated assistance that still needs validation for consequential decisions.
Troubleshooting signals
- Confirm the user’s license or metered access, the experience’s availability, source permission, agent audience, and whether the chosen tool matches the task.
- If an agent response lacks expected knowledge, test the source directly, review selected knowledge and instructions, and distinguish retrieval failure from unclear prompting.
- If an action fails, inspect connector authentication/permission and action configuration rather than changing knowledge-source access.
More detail
- Standard Microsoft 365 Copilot is the general-purpose experience for drafting, summarizing, asking questions, preparing meetings, and working in Microsoft 365 apps. It is optimized for broad everyday tasks rather than one fixed departmental workflow.
- Researcher is a Microsoft first-party agent for complex, multi-step research. It can draw from permitted work content and the web, takes longer than ordinary chat, asks clarifying questions, and returns a structured, source-cited report suited to decisions and presentations.
- Analyst is a Microsoft first-party agent for advanced data analysis and reasoning. It is appropriate for exploring data, calculations, trends, anomalies, and visualized findings rather than conducting a broad literature-style report.
- Microsoft also offers prebuilt agents for recognizable roles or tasks. Names and availability can evolve; the durable exam decision is whether a Microsoft-provided specialist fits the need or the organization must build and govern its own agent.
- A SharePoint agent is grounded in selected SharePoint sites, libraries, folders, or files and answers within each user’s permitted source access. It is useful for a focused knowledge experience close to team content.
- A custom agent packages a purpose, owner, audience, instructions, knowledge, suggested prompts, and—only when needed—connectors or actions. Lightweight agents serve focused Q&A; advanced agents can integrate systems and multi-step workflows.
- Copilot and agents share Microsoft 365 identity, permissions, security, and compliance foundations. An agent does not gain unrestricted data or action rights merely because it is specialized.
Ready for the quiz?
- When is standard Copilot enough?
- What makes Researcher different from Analyst?
- Why is a saved prompt not an agent?
- When is a SharePoint agent the focused choice?
- Which requirements justify building a custom agent?
Related objectives
- D3.1.a — Compare the built-in capabilities of Copilot and agents
- D3.1.d — Identify use cases for Researcher
- D3.1.e — Identify use cases for Analyst
- D3.1.f — Identify use cases for custom agents