– The estimated reading time is 6 min.
Imagine an office worker clicking on a phishing link at 2:13 a.m. As the IT team sleeps, an AI agent detects unusual login behavior and disables the compromised account. Then, it checks to see if malware spread to any other devices or accounts, opens a ticket, sends an alert to the security team and drafts an incident summary.
This is the beauty of AI agents. They can help workers in all kinds of roles do a range of things, like track project deadlines, monitor shared inboxes, create nightly reports and more. What sets them apart from AI chat apps is that while chat apps excel at answering questions, agents can also take action.
You can think of agents as apps for an AI-powered world. Getting started is easier than it sounds. This guide walks through the key steps — from defining the problem to building and sharing an agent using Microsoft 365 Copilot — so you can go from idea to working prototype without writing any code.
If you have Microsoft 365 Copilot, go to [Microsoft 365 Copilot Chat], click “Agents” and then “New Agent” to start building immediately.Or you can learn more about
[Copilot and AI agents]. **1. Start with the problem **
Begin with the job, not the technology. What problem are you trying to solve? Talk with colleagues to clarify the need and narrow the scope. Then define the outcome: Should the agent retrieve information, complete a task or act independently? A clear goal upfront makes the rest of the build process faster — and far more useful.
Example: A group of office workers spends hours each week pulling together a status report from their emails, messages and documents. It’s repetitive and easy to miss details, and the final report varies widely depending on who puts it together. An AI agent could solve that by doing the legwork of reading the team’s internet strata and putting together the first draft in minutes.
2. Decide where to begin
Before building, check what already exists. An AI model or prebuilt agent may already do the job. To check, open Microsoft 365 Copilot and click on “Agents” in the left panel to open a store of prebuilt agents. If you can’t find exactly what you’re looking for, consider your level of expertise. With little or no coding experience, Microsoft 365 Copilot is designed to help you get started quickly. More advanced builds may require developer tools for greater customization and control.
Example: For a team that works together to respond to client emails from a shared inbox, an agent could help sort incoming messages, pick out which are most important, write replies for common questions and route messages to the right person.
3. Build your AI agent in Microsoft 365 Copilot
From the Agents section in Microsoft 365 Copilot, click “New Agent.” Start by describing what you want in plain language — your copilot will generate a draft of your agent for you. Once Copilot has built your agent, you can test it by clicking “configure.” Then refine the instructions: Define the agent’s behavior, tone and tasks. This step is where you shape how the agent responds, ensuring it’s consistent, useful and aligned with your specific workflow.
Example: The team responding to emails in a shared inbox could give its agent all kinds of direction. A team member could tell it to categorize common request types as general questions, urgent issues or complex requests, and to route each type of message to a specific team member. The team could train the agent to use certain approved language to respond to routine questions right away and to flag anything sensitive to a human reviewer.
4. Add knowledge and define outputs
Getting your agent to do its best work requires you to give it a little context.
Connect your agent to the right information sources, such as emails, documents, SharePoint sites or websites. Decide whether it should rely only on your curated data or pull from broader sources, like a website or a PDF of your company’s policies. Then define what it produces — reports, presentations, spreadsheets, responses or code — so it delivers work in the format you actually need.
You can simply attach documents to your agent yourself or use the chatbot. To do this, navigate to the Agent Builder page in your Microsoft 365 Copilot, and click the pencil icon to the right of your agent’s name to open the edit screen. From there, you can scroll down to attach information sources or use the chat window on the left to walk you through it.
Example: The team building a weekly report could tell the agent to “review Teams messages and emails from the past seven days” and to “focus only on messages related to a specific project or team,” or to specifically “identify updates, decisions, blockers or upcoming deadlines.”
Once that’s done, the team can tell the agent exactly how it wants the report to be structured and how long it should be, and to write the whole thing in a clear, professional tone. The team can also set boundaries with the agent and tell it to only use information from emails and Teams messages — and not to infer any missing details.
5. Test, share and scale
Test your agent in real scenarios and refine its instructions by hopping back into the settings as needed. From the edit screen within the Agent Builder page, you can change the agent’s instructions yourself, or Copilot can walk you through changes on the left in a chat window.
Example: The team with the weekly report may realize its agent is getting unclear or conflicting updates from the team’s emails, so a team member instructs the agent to flag these — and not to infer or guess any missing details.
Or the team might realize its initial word count for the report was too long, so a team member adjusts the agent’s directions to make it shorter and more skimmable, or to include key points in a bulleted list.
Once your agent is up and running, you can expand its capabilities or move it into more advanced tools. The goal isn’t perfection right away — it’s building something useful for your work and continuing to refine it over time.
All images made using MAI Playground and Microsoft 365 Copilot.
*Samantha Kubota reports on everything AI and innovation for Microsoft Signal, with a recent focus on how AI agents are reshaping everyday work, Microsoft’s research breakthroughs and the responsible use of emerging technologies. Prior to Microsoft, she was a journalist at NBC News. Follow her on LinkedIn. *