Chapter 8 — Building and Using Copilot Agents¶

Part II — AB-730 track: Working with Microsoft 365 Copilot

In 30 seconds¶

The core idea: agents are purpose-built Copilots. You can get one from the Agent Store or build your own from a template, give it knowledge, configure its instructions / capabilities / suggested prompts, and share it — all with no code.
Why it matters: agents are a large, explicit part of AB-730 Domain 2.
The exam angle: expect questions on Agent Store vs building your own, and on each configuration element.
Remember: use an existing agent when one fits; build one when you need specific knowledge and consistent, reusable behavior.

Exam map¶

Exam map — AB-730 · Domain 2: Create and manage Microsoft 365 Copilot agents

1. Key concepts¶

📖 Definition — Agent: a customized assistant built on Microsoft 365 Copilot, scoped to a specific job with its own instructions, knowledge, and suggested prompts. Agents range from simple Q&A helpers to more autonomous task performers.

📖 Definition — Agent Store: the catalog where you find and add ready-made agents — from Microsoft, partners, and your own organization — without building anything.

📖 Definition — Copilot agent builder: the no-code tool for creating a declarative agent by describing what it should do, giving it knowledge, and configuring its settings.

Agent Store vs creating your own¶

flowchart TD
    A["You have a repeatable need"] --> B{"Does an existing agent<br/>already fit?"}
    B -->|Yes| C["Use it from the Agent Store"]
    B -->|No| D["Build your own with a template<br/>(add knowledge + settings)"]
    D --> E["Share it with your team"]

📌 Key concept: don't build what already exists. Check the Agent Store first; create a new agent when your need is specific to your content and process and no existing agent covers it.

🎯 Exam tip: "understand when to use Agent Store versus creating a new agent" is stated verbatim in the objectives. Existing/general need → Agent Store. Specific knowledge + consistent behavior + reuse → build your own.

2. How it works — building and configuring an agent¶

You create an agent from a template in the agent builder, then configure four things:

Setting	What it controls	Example
Instructions	How the agent behaves, its tone, its rules	"Answer HR policy questions; cite the policy; be concise"
Knowledge	The sources it grounds on	A SharePoint site of HR policies
Capabilities	Extra abilities you enable	Web search, image generation, code interpreter
Suggested prompts	Starter prompts shown to users	"How many vacation days do I get?"

🔍 How it works: an agent grounds on the knowledge you attach (for example, a SharePoint site), respecting each user's permissions — the same boundary as Chapter 2. Its instructions shape every response so behavior is consistent for everyone who uses it.

💡 Tip: good agent instructions read like a clear job description — what it does, what sources to use, what tone, and what it should not do. This is prompt engineering (Chapter 3) applied to configuration.

Once built, share the agent with teammates so everyone gets the same purpose-built help. As always, each user's results stay permission-trimmed to what they can access.

⚠️ Pitfall: sharing an agent doesn't grant users access to its knowledge sources they couldn't otherwise see. If a teammate lacks permission to the underlying SharePoint site, the agent won't surface that content to them.

3. In the real world¶

Scenario — the HR helpdesk agent. HR is flooded with the same policy questions. Instead of answering each one, an HR lead opens the agent builder, starts from a template, writes instructions ("answer employee questions using our official HR policies, cite the source, stay concise"), attaches the HR policy SharePoint site as knowledge, adds suggested prompts ("How do I request parental leave?"), and shares it with all staff. Employees get instant, consistent, cited answers — and each person only sees policies they're allowed to. No code, no new headcount.

4. Exam tips¶

🎯 Exam tip: know the four configuration elements — instructions, knowledge, capabilities, suggested prompts — and what each does.

🎯 Exam tip: agents are built without code (the agent builder is no-code). Answers implying you must program an agent are wrong for these business exams.

🎯 Exam tip: an agent's knowledge respects permissions. Sharing an agent shares the assistant, not access to its underlying data.

5. Common pitfalls¶

⚠️ Pitfall: building a new agent when a suitable one already exists in the Agent Store — check first.

Vague instructions: an agent with weak instructions behaves inconsistently; be explicit about scope, sources, and tone.
Wrong or missing knowledge: without the right knowledge source, an agent can't ground answers in your content.
Assuming sharing grants data access: it doesn't — permissions to knowledge sources still apply.
Thinking agents require coding: the agent builder is no-code.

6. Practice questions¶

1. A team needs an assistant that answers questions from their internal product documentation, consistently, for everyone. No suitable agent exists. What's the best approach?

A. Tell everyone to use general chat and re-explain context each time
B. Build a new agent, attach the product docs as knowledge, configure instructions, and share it
C. Email the documentation to everyone
D. Fine-tune a custom language model

Answer

Correct: B. A specific, repeatable, knowledge-scoped need with no existing agent is exactly when to build one. A is inefficient; C isn't AI-assisted; D is unnecessary and out of scope (no code / no fine- tuning needed).

2. Which agent setting determines the sources an agent uses to ground its answers?

A. Suggested prompts
B. Capabilities
C. Knowledge
D. The agent's name

Answer

Correct: C. Knowledge defines the grounding sources (e.g., a SharePoint site). Suggested prompts are starter questions; capabilities add abilities like web search; the name is just a label.

3. When should you use the Agent Store instead of building your own agent?

A. Never — always build from scratch
B. When an existing agent already meets the need
C. Only when you can write code
D. Only for personal use

Answer

Correct: B. If a ready-made agent fits, use it rather than rebuilding. A wastes effort; C and D are irrelevant to the choice.

4. A shared HR agent uses a SharePoint site as its knowledge. A contractor without access to that site uses the agent. What happens?

A. The agent surfaces the restricted content anyway
B. The agent won't surface content the contractor lacks permission to see
C. The agent grants the contractor access to the site
D. The agent stops working for everyone

Answer

Correct: B. Agents honor each user's permissions; sharing the agent doesn't grant data access. A and C violate the security model; D is false.

5. Which set correctly lists agent configuration options in Microsoft 365 Copilot?

A. Instructions, knowledge, capabilities, suggested prompts
B. CPU, memory, disk, network
C. Fonts, colors, themes, icons
D. Tokens, weights, layers, epochs

Answer

Correct: A. Those are the agent-builder settings. B is infrastructure; C is cosmetic; D are model-training internals irrelevant to configuring a no-code agent.

Chapter 8 — Building and Using Copilot Agents¶

In 30 seconds¶

Exam map¶

1. Key concepts¶

Agent Store vs creating your own¶

2. How it works — building and configuring an agent¶

Sharing an agent¶

3. In the real world¶

4. Exam tips¶

5. Common pitfalls¶

6. Practice questions¶

Further reading¶