Imagine giving two employees the same task, but one has all the right tools and the other doesn’t. You’d get very different results. I experienced the same with Microsoft’s Copilot Studio agents – and learned that how you build your AI assistant makes a huge difference in what answers you get from SharePoint.
Choosing the right kind of Microsoft Copilot Studio agent can make a big difference in how well you get answers from your SharePoint data (I’ve been battling the Copilot Studio Agents myself for months and seen the results firsthand). Microsoft’s Copilot Studio lets you build AI assistants (agents) to help with tasks or questions. But not all Copilot agents are created equal, especially when it comes to searching your SharePoint documents and sites. In this article, we’ll explain what Copilot agents are, why the type of agent you choose matters for your business, and what you should do next to get the most out of this technology.
What is it?
Microsoft Copilot agents are AI assistants you create for your organization’s needs. Copilot Studio offers two ways to build one, and the choice matters: a Custom Copilot Agent or an M365 Copilot Agent. In simple terms, one is your custom-built AI (you configure its “brain” largely yourself), and the other uses Microsoft’s built-in AI brain with just a bit of tuning from you. When I tried both, the Microsoft 365 (M365) Agent effortlessly pulled up relevant documents from our SharePoint, while the Custom Agent often came up short-handed with much fewer results. Microsoft actually expects this (Response Parity) – these two agents run on different underlying technology – so it’s important to choose the right type for your business scenario.
- Copilot Studio Custom Agent: This is a bit more of a do-it-yourself AI agent (still easy to setup within the CS no-code environment). Think of it as building a custom chatbot from the ground up (with Microsoft’s no-code tools to help you). It’s powerful because you can tailor it heavily. For example, you might integrate it with a third-party database or give it a unique decision-making flow.
- Copilot Studio M365 Agent: This agent is more like a pre-trained AI assistant that you customise. It runs on the same technology that Microsoft 365 Copilot uses out-of-the-box. In practice, that means an M365 Agent already knows how to search your Microsoft 365 data (in this case SharePoint). You just point it to the specific sites or files it should focus on, and it leverages the full Microsoft AI engine behind the scenes. It’s like having a seasoned employee who already understands the company’s intranet versus a new hire who needs guidance – the seasoned one (M365 Agent) can find information much more accurately.
In short: A Custom Agent gives you a bit more control and flexibility (still in the no code/low code world), while an M365 Agent comes with Microsoft’s handles Microsoft 365 content much better. Both are built in Copilot Studio, but they provide vastly different results when accessing SharePoint data (and behind the scenes use different orchestration engines).
What does it mean from a business perspective?
From a business point of view, the choice between a declarative M365 Copilot agent and a custom Copilot Studio agent can have real impacts on trust, productivity and outcomes. Here are the key implications:
- Relevance of Answers: Agents built on the Microsoft 365 Copilot retrieve more relevant and accurate information from SharePoint. In contrast, a custom-built agent may struggle to find what you’re looking for in SharePoint data, even if that data exists. This leads to incomplete answers. (In my experience the M365 Agent was simply superior when accessing SharePoint data.)
- Trust from Consistency: The M365 Copilot agent provides responses that are well-formatted and phrased in a helpful tone, and importantly from a trust perspective – much more accurate. Businesses benefit from this consistency – employees get answers that feel like they come from a knowledgeable assistant.
- Productivity Impact: Because a declarative Copilot agent finds data more effectively, employees can get their answers faster and with more confidence.
- Maintenance and Development: Using a declarative agent is easier to set up and maintain. You configure it with instructions and connected data, and Microsoft’s system handles the AI tuning, updates, and any improvements over time.
What do I do with it?
If you’re looking to implement a Copilot agent for your business, here are some concrete steps and recommendations:
- Match the agent type to your scenario: Evaluate what you need the AI agent to do. If your goal is to answer employees’ questions using content on your SharePoint or Office 365 (and you want the best search results with minimal tweaking), start with a Microsoft 365 Copilot agent.
- Pilot and compare performance: It’s a good idea to do a side-by-side pilot. Build a simple prototype of a M365 Copilot agent and a custom agent addressing the same use case, then ask them typical questions and observe the quality of answers. For example, do a test where both agents are connected to the same SharePoint documents and ask, for example “Provide a summary of subject X?” or “Identify all documents and summarise them on subject X.” This will quickly show you the difference in how each one retrieves and presents info.
- Consider licensing and rollout strategy: The declarative Copilot agents require user licenses. Microsoft 365 Copilot is a premium add-on, so not every organization or team within an organization will have it for everyone – check with your technology teams on the implications of using the M365 Agent and who can access it.
- Monitor performance and iterate: Treat your AI agent like a living project (Microsoft is continually enhancing the environment so this behaviour may change over time), not a one-time installation. Monitor what questions are being asked and where the agent falls short and repeat these tests as you modify the agent.
- Stay informed on updates: The AI landscape (especially Microsoft’s Copilot ecosystem) is evolving quickly. Microsoft is likely to bring improvements to Copilot Studio’s custom agents (closing the quality gap) and introduce new features for declarative agents.
How will you leverage Copilot agents in your organization? Start with a small pilot (choose the right agent for the job), gather feedback from your team, and iterate. By being deliberate in how you build and roll out these agents, you can turn AI into a trusted co-worker that makes everyone’s job easier.
Let’s navigate this new era of AI Agent-powered work together and make sure our tools are actually helping us get the job done better – share your perspectives in the comments!
Further Reading
Breaking Down Copilot Agents (Steve Corey)