Manage prompts and conversations by using AI
Create, Configure, and Share Copilot Agents
CoreReuse or build the right agent, then separate templates, knowledge, instructions, capabilities, suggested prompts, and sharing roles.
Aligned to the AB-730 skills measured as of July 22, 2026; product behavior verified August 23, 2026.
Why this matters
A useful business agent is not defined by its name. Its source grounding, behavioral instructions, supported functions, onboarding prompts, testing, and audience must each satisfy a distinct requirement.
Must Know
- Agent Store is the discovery hub for available agents from Microsoft, trusted partners, and the organization. Evaluate an existing agent against the business need before creating a duplicate.
- Use a template when its scenario and preconfigured elements are a suitable starting point, then tailor its name, description, instructions, knowledge, capabilities, and suggested prompts as needed.
- Knowledge supplies authoritative facts and content the agent should draw from. Choose current, approved, relevant sources and respect their permissions.
- Instructions define role, scope, workflow, boundaries, tone, and response behavior. Do not use instructions as a substitute for authoritative reference material.
- Capabilities add supported functions. Suggested prompts help users begin appropriate tasks; they do not control the agent’s underlying behavior.
- Test the agent with realistic and boundary cases before sharing. Diagnose whether a bad response is a knowledge, instruction, capability, or prompt-onboarding problem.
- A newly created Agent Builder agent is private to its creator. Share with intended users as Can chat when they should use but not edit it; grant Can edit only to actual co-owners.
- Sharing an agent does not automatically bypass the recipient’s access to underlying knowledge and does not require making every recipient an owner.
Compare and Distinguish
- Agent Store vs create: use a suitable available agent vs build for an unmet tailored requirement.
- Template vs blank creation: start from relevant preconfiguration vs define the experience from the beginning.
- Knowledge vs instructions: facts and grounding vs rules for role and response behavior.
- Capability vs suggested prompt: a function the agent can perform vs an example starting request for users.
- Can chat vs Can edit: interact with the agent only vs own, edit, share, and manage it.
- Researcher vs Analyst: multi-source research and sourced reporting vs quantitative data exploration and insights.
Scenario examples
- Scenario: An approved agent already performs market research with the required sources. Think: use it from Agent Store instead of rebuilding.
- Scenario: An onboarding template is close but generic. Think: tailor its instructions and approved handbook knowledge before testing.
- Scenario: The agent knows policy facts but answers outside scope and with the wrong tone. Think: change instructions, not knowledge.
- Scenario: Teammates should ask the agent questions but must not alter it. Think: share with Can chat and verify their underlying source access.
Exam traps
- Topical similarity in Agent Store does not prove an agent meets required knowledge, behavior, privacy, or audience constraints.
- Selecting a template does not finish configuration.
- Putting reference facts into instructions blurs grounding and makes maintenance harder.
- Suggested prompts do not grant capabilities, and capabilities do not explain how the agent should behave.
- A chat link or agent share does not grant source-file permission; Can edit is broader than ordinary use.
Key takeaways
- Reuse when fit is real; create when tailored requirements remain unmet.
- Knowledge grounds, instructions guide, capabilities enable, and suggested prompts onboard.
- Test before sharing, choose the narrow audience and role, and preserve source permissions.
How it works
- The builder combines the selected template or description with configured knowledge and behavior. The creator tests outputs, corrects the responsible setting, creates the agent, and shares it deliberately.
- At use time, the agent applies its configuration while the user’s access and the organization’s controls continue to govern source content.
Objects and administrative surfaces
- Agent Store — discover and install or use agents made available to the user.
- Agent Builder in Microsoft Copilot — natural-language and form-based creation, configuration, testing, and sharing for straightforward agents.
- Templates — preconfigured starting points that still require fit review and tailoring.
- Share dialog — assign Can chat or Can edit to named users or groups as supported; organizational policy can constrain available sharing choices.
When to use it
- Use a template for a close scenario match that benefits from a faster starting point.
- Add knowledge for approved facts, instructions for repeatable behavior, capabilities for needed functions, and suggested prompts for common starting tasks.
- Share Can chat with intended users; add individual Can edit owners only when shared maintenance is required.
Security and governance implications
- Keep knowledge scoped to necessary approved material and confirm intended users can legitimately access it.
- Avoid org-wide or broad sharing when the agent’s purpose or knowledge is team-specific.
How to reason about this
- Wrong facts suggest a knowledge source or grounding problem; wrong tone or process suggests instructions; missing function suggests capabilities; users starting the wrong tasks suggests suggested prompts.
- If a recipient can open the agent but not retrieve expected content, verify their source permission and required capability availability before expanding sharing.
More detail
- Agent Builder supports dedicated SharePoint and connector-based knowledge for straightforward scenario-specific agents, plus an interactive test experience before use or sharing.
- Agent sharing can assign chat users and owners. Owners have equal management rights; a group can be a chat user but cannot be added as an owner in the current sharing model.
- Organizational policy and product availability can constrain creation or sharing. AB-730 tests the user-visible choice, not administrative deployment mechanics.
Ready for the quiz?
- What requirement belongs in knowledge rather than instructions?
- When is a template the right starting point?
- Which setting fixes a missing supported function?
- Which sharing role fits coworkers who should use but not edit the agent?
Related objectives
- D2.3.a — Understand when to use Agent Store versus creating a new agent
- D2.3.b — Create an agent by using a template
- D2.3.c — Configure an agent that has knowledge
- D2.3.d — Configure agent settings such as instructions, capabilities and suggested prompts
- D2.3.e — Share an agent with team members