Copilot in Excel started as your built-in AI assistant, something you could chat with right inside your workbook. You could ask natural language questions like “summarize sales by region” or “find trends in this data,” and Copilot would instantly generate the right formulas, charts, or summaries to help you explore your data.
Now we’re entering what Microsoft calls the agentic era, a new stage where AI doesn’t just assist you but begins to take action. Inside Excel, this means you can call on specialized agents, pre-built and task-focused versions of Copilot that can reason about your data, gather context, and act on your behalf.
To see what this looks like in practice, we’ll use a dataset on global CO₂ emissions. Within Excel, two pre-built agents available for Microsoft 365, Researcher and Analyst, offer different perspectives on this data. The Researcher agent helps you understand context and background, while the Analyst agent focuses on computation and insights.
Together they show how Copilot is evolving from a simple assistant into a true reasoning partner for your data. Let’s see it in action. Download the exercise file below and follow along.
To find these agents in Excel, open the Copilot button and select Copilot Chat. Then, toggle the navigation pane in the upper-left corner of the window. You’ll see a list of agents, some preloaded by Microsoft, others that you can sync or even create yourself. You’ll also notice that Copilot Chat is listed among them. The implication here is that Copilot is just one of many agents you can work with.
Researcher Agent: context and framing
Unlike the traditional Copilot in Excel experience you might be used to, Researcher doesn’t interact directly with your workbook data. Instead, it draws on trusted sources and knowledge models to help you shape and frame your analysis.
You might start by asking Researcher:
“Summarize global CO₂ emission trends over the last 50 years.”
You can even choose what kinds of sources Researcher can draw from to provide context, whether that’s information from the web, internal company data, specific SharePoint sites or more.
After a few moments, Researcher produces a neatly formatted, report-style summary that brings together all the information sources it can access… including, yes, the data in your workbook.
So even though it doesn’t interact directly with your data in the same way Copilot does, it can still read it to inform its analysis. And because our data was stored in a properly structured Excel table with clear column headers, Copilot could understand it much more easily and generate cleaner, more relevant results.
Let’s look at one more example. The goal here again is to think broadly about how you’re approaching your research. Consider what kinds of context you might want to include, what questions are worth exploring, and what insights you might want to explain or connect back to the data itself.
“Explain how CO₂ emissions relate to GDP and industrialization.”
In other words, think about the bigger story of how you came across this data, why it matters, and what perspective you’re trying to uncover through your analysis. Rather than performing calculations, you’re using Researcher to think through the problem, define your approach, and clarify what matters.
In short, the researcher helps you focus on the “why” behind your work, giving you the background and conceptual grounding to make your data exploration meaningful. What makes this an agent is that it operates with a defined purpose and set of capabilities. It doesn’t just respond to a single question and stop there. It reasons across multiple sources, builds structured outputs like reports or summaries, and adapts its behavior based on your prompts and settings.
Analyst Agent: Data-driven execution
Once you’ve set the direction, the next step is analysis. The Analyst agent operates differently from Researcher because it works directly with your Excel data. You can also upload additional datasets if you want to expand its scope. This is where the idea of an agent really starts to come to life. Analyst can plan, reason, and carry out a sequence of steps to accomplish a defined goal.
You might ask Analyst to do things like:
“Find the top five countries with the highest CO₂ emissions in the most recent year.”
Notice that Analyst doesn’t search the web or pull in outside information. Its focus is entirely on the data at hand, making it ideal for in-depth, workbook-level analysis. In many ways, it operates like the Advanced Analysis feature of Copilot, where you can ask to perform more complex or custom computations.
If you open the dropdown menu while it’s running, you’ll see exactly what’s happening behind the scenes: much like Advanced Analysis, Python code is being generated and executed to carry out the analysis. This gives you transparency into how the results are produced and lets you learn from or even customize the underlying logic if you want to refine your approach further.
Please note that there’s a lot happening behind the scenes here: reasoning loops, code checks, and data validations running in sequence. It may take a little time to process, so be patient while the agent works through each step to deliver accurate and reliable results.
Next, I’ll ask for a data visualization:
“Create a chart showing total CO₂ emissions over time for the top three countries.”
Not only do I get a clean, well-formatted chart in return, but it even includes interactive tooltips. This visualization is created using a Python library called Plotly, which unfortunately isn’t yet supported natively in Python in Excel.
However, the underlying code is fully visible, so you can copy it, save the image, or adapt it as needed. With a bit of help from Copilot, it wouldn’t take much to refactor this Plotly chart into a native Python in Excel visualization.
What makes Analyst an agent is that it doesn’t just execute a single command or calculation. It operates with intent and autonomy toward a defined analytical goal. It plans a sequence of reasoning steps, checks its own work, and adapts as it goes, much like a skilled analyst would. Rather than returning one-off results, it manages a full analytical workflow: identifying the right computations, generating Python code, validating outcomes, and presenting insights in clear, visual form.
Conclusion
Researcher gives you context. Analyst gives you execution. Together, they show how Copilot in Excel is moving from a simple helper to an intelligent partner that can reason, plan, and act. This builds on the basic Copilot experience. Where Copilot Chat handled single questions or formulas, agents like Researcher and Analyst can connect steps, hold context, and adapt to your goals.
Tools like Copilot Studio and Power Automate will only extend this further, letting you design full workflows that link multiple agents and data sources. Together, they form a growing Excel AI stack built for reasoning, automation, and insight:
Thanks for exploring this with me. I hope it helps you see how these new agentic tools can take your Excel work to the next level and inspire new ways to analyze, automate, and create with your data.