Building your first Copilot agent
1. Building your first Copilot agent
In this video you'll build your own agent in Agent Builder. We'll make a bot that answers questions about a product launch.2. What is Agent Builder?
So, first: what is Agent Builder? It's a guided tool inside the Microsoft Copilot app. It lets you create a custom Copilot that understands your business content and answers questions based on it. The result is a specialist Copilot, focused on a single job or domain, such as handling FAQs or analyzing product launch materials. There are two ways in. You can describe the agent you want in plain language and let Copilot draft it, or set it up field by field. We'll do it manually, because seeing the fields is what teaches you how an agent works.3. Where Agent Builder is available
Agent Builder isn't in every Microsoft 365 plan - it's in the enterprise-level versions of Microsoft 365 Copilot. On a personal or small business plan you may not see it yet. Where it is available, it lets you shape Copilot into a dedicated expert for your team, built around the work you actually do.4. What's involved in building an agent
Before you start building, it helps to understand what's involved. In Agent Builder, an agent is really three fields: First, a name that reflects its role or purpose.5. What's involved in building an agent
Second, instructions that explain what it should do and how it should answer. This is the one field that really shapes its behavior.6. What's involved in building an agent
And third, a knowledge source, like a SharePoint site or a document library, that provides the information it will rely on to answer questions. Name, instructions, knowledge. Everything else in the Configure pane is optional.7. Plan your agent with GCSE
The GCSE framework helps you define your agent with clarity and purpose: Goal: What do you want the agent to do? Context: Who will use it and why? Source: What content should it rely on? For example, a SharePoint site. Expectation: How should responses be formatted? Defining an agent has a lot in common with writing a job description for a new team member, and the clearer you are, the better the result.8. Plan your agent with GCSE
A Copilot chat is used once. An agent is persistent - people come back to it again and again. That's why the Goal and the Source have to be clear. This is where an agent earns its place. If thirty colleagues each ask you the same five questions about the launch, that's a hundred and fifty answers you write by hand. Build the agent once, point it at the launch documents, and it answers them instead, from the same source every time. A quick tip - use the GCSE framework to structure your agent's instruction set. This gives you a repeatable system that helps your agent stay focused, relevant, and reliable.9. Describe your agent
Open Agent Builder and give your agent a name and instructions. Name it "Launch FAQ Helper", so its purpose is obvious to anyone who finds it. Then the instructions: "You are the EcoGreen Launch FAQ Helper. Answer employee questions about the EcoGreen reusable water bottle launch: its benefits, timeline, and team responsibilities. Keep answers short and in bullet points, and base them only on the uploaded launch documents." Those few lines tell the agent who it is, what it covers, how to format answers, and what to base them on.10. Add a knowledge source
Click Add knowledge, then paste the SharePoint URL for your launch documentation. Start with just one trusted source. A single authoritative source keeps answers accurate and reduces the risk of conflicting ones. You can always expand to more files or sites later.11. Test and refine
Now it's time to test your agent. Use the Preview panel to ask a real question, for example: "What are the top 3 benefits of the EcoGreen reusable water bottle for customers?" If the answer doesn't look right, adjust the instructions or update the knowledge link and test again. You're not locked into the first version.12. Let's practice!
Each refinement makes the agent more reliable for the people using it. Now, it's time for you to try building your own agent in the exercises ahead!Create Your Free Account
or
By continuing, you accept our Terms of Use, our Privacy Policy and that your data is stored in the USA.