What Are Agents?
Agents are AI assistants you can customise for specific jobs. Think of them as Copilot's specialist colleagues — each one built for a particular task or dataset.
What’s an agent, and how is it different from Copilot?
Think of Copilot as a clever generalist — and agents as specialists you hire for specific jobs.
Copilot is like a smart colleague who can help with anything — write an email, summarise a meeting, crunch some numbers. It sits inside your apps and waits for you to ask.
An agent is different. An agent is built for ONE specific purpose. Maybe it answers questions about your company’s product catalogue. Maybe it processes new employee onboarding. Maybe it monitors your SharePoint site and surfaces the right documents to visitors.
The key difference: Copilot assists you. Agents can work alongside you — or even on your behalf.
Four types of agents
Not all agents are the same. Microsoft organises them into four categories — and the exam tests whether you know the difference.
| Feature | Who builds it | Complexity | Example |
|---|---|---|---|
| Prebuilt agents | Microsoft or approved vendors | Ready to use, no setup | Prompt Coach, Writing Coach, Researcher, Analyst |
| SharePoint site agents | Auto-created by M365 Copilot | Ready-made for each site | A site agent that answers questions about documents on that site |
| Everyday business user agents | Anyone (no coding needed) | Simple, guided creation | An HR agent that answers onboarding questions from a SharePoint library |
| Advanced agents | Developers | Custom connectors, generative AI, APIs | An invoice processing agent that connects to your ERP system |
Prebuilt agents you should know for the exam
Microsoft provides several prebuilt agents. The exam expects you to recognise these:
- Prompt Coach — helps you write better prompts for Copilot
- Writing Coach — gives constructive feedback on any writing task
- Researcher — gathers, analyses, and synthesises information from your work data AND the web into structured reports
- Analyst — transforms raw data into insights and visualisations using advanced reasoning and Python
Exam tip: Researcher works with unstructured knowledge (documents, emails, web pages). Analyst works with structured data (spreadsheets, databases). Know which one to recommend for a given scenario.
How agents get their knowledge
An agent is only as good as the data it can access. When you create an agent, you connect it to knowledge sources — these determine what the agent knows and can answer questions about.
| Knowledge Source | What It Provides | Example |
|---|---|---|
| SharePoint content | Internal documents, policies, procedures | An HR agent grounded in the company handbook |
| Web content | Public websites and external information | A marketing agent that references your public product pages |
| Copilot connectors | External apps like Salesforce, Jira, ServiceNow | A helpdesk agent that pulls open tickets from ServiceNow |
Real-world scenario: Brew & Byte's first agent
Brew & Byte (30-person creative agency) needs a simple way for new hires to find answers about company policies — dress code, expense claims, holiday requests.
Solution: Kai (the owner) creates an agent in Copilot Chat:
- Opens Copilot Chat → clicks “Create agent”
- Describes the agent: “An assistant that helps new employees find answers about company policies”
- Connects it to the SharePoint site where HR documents live
- Shares it with the team
Total time: 15 minutes, zero code. New hires ask the agent “How do I claim expenses?” and it pulls the answer from the company’s expense policy document.
This is an everyday business user agent — no developers needed.
Two tools for creating agents
| Tool | Who uses it | Creates agents for |
|---|---|---|
| Copilot Studio | Business users (lite mode) AND developers (full mode) | Copilot Chat agents |
| Agent tool in SharePoint | Business users (no coding) | SharePoint site agents |
Key distinction for the exam: Copilot Studio has two experiences:
- Lite experience — simple, guided, for everyday business users (no code)
- Full experience — advanced features, custom connectors, generative AI (for developers)
Both tools capture similar information (name, description, instructions, knowledge sources), but Copilot Studio supports additional capabilities like code interpreter and image generator that aren’t available for SharePoint agents.
SharePoint’s ready-made agents
Every new SharePoint site can get a ready-made agent automatically created by Microsoft 365 Copilot. This agent is scoped to the content within that site — it can answer questions about documents, pages, and files on the site.
Important details for the exam:
- The ready-made agent becomes the site’s default agent
- Site owners can replace it with a custom agent later
- The agent respects user permissions — it only shows content the user can access
- Not every site gets one — it depends on licensing and configuration
🎬 Video walkthrough
🎬 Video coming soon
What Are Agents? — AB-900 Module 22
What Are Agents? — AB-900 Module 22
~10 minFlashcards
Knowledge Check
Brew & Byte wants to create an agent that helps clients find information about their design services. The agent should pull answers from both their public website and internal SharePoint documents. Who should build this agent?
Northwave's SharePoint site for the legal department was just created. Maya notices there's already an agent available on the site. Which type of agent is this?
Next up: Copilot vs Agents — a deeper comparison of when to use each, with real-world scenarios to help you choose the right tool for the job.