Agent Mode in Excel is a Powerhouse
Turn Excel Copilot into a hands-free data analyst.
Hey friends 👋
Welcome to another Signal Pro workflow issue.
Excel’s Agent Mode quietly rolled out to general availability across web, Windows, and Mac back in January. Most people I’ve spoken to either haven’t heard of it or tried regular Copilot once, got a disappointing result, and moved on.
Agent Mode is completely different.
Regular Copilot chat answers questions about your data. Agent Mode actually builds and edits your workbook directly. This includes creating sheets, writing formulas, generating charts, cleaning data, and building PivotTables, all done step by step, while you watch.
It’s a bit like having a very smart colleague sitting next to you.
Today I’ll walk you through:
How to set up Agent Mode (it takes 30 seconds)
Building a complete dashboard from an empty workbook using nothing but conversation
A skill sheet technique that makes every chart Copilot creates match your brand automatically on the first pass
Let’s get into it.
What is Agent Mode?
First, let’s be clear about what this actually is. Agent Mode is a setting inside Copilot in Excel that switches the AI from a Q&A assistant into an autonomous agent.
Instead of answering one question at a time, it takes your goal, breaks it into steps, executes each step directly in your workbook, and validates the results before moving on. You can see its reasoning in the chat pane as it works, which is one of my favourite things about it.
It works on Excel for the web, Excel for Windows, and Excel for Mac. You can choose between OpenAI’s GPT 5.2 and Anthropic’s Claude Opus 4.5 as the underlying model with a model picker built right in.
A few things to know before you start:
Who has access: You need a Microsoft 365 Copilot commercial license, or a Microsoft 365 Personal/Family/Premium subscription (Personal and Family use AI credits). If you’re on a business plan and can’t find it, your IT admin may need to enable it.
What it can do: Create and format worksheets, write formulas, build charts and PivotTables, clean messy data, search the web for current information, and iterate on any of the above through follow-up prompts. You can also attach files, add work content, and pull in web sources directly through the chat. This is especially handy if you’re working with 10-K filings or external data.
What it can’t do: Access other files, emails, or external enterprise data. It only works with the workbook you have open.
Right. Let’s set it up.
Setting up Agent Mode
This takes about 30 seconds.
Step 1: Open Excel for the web. I’m using the web interface for today’s demo. Agent Mode also works on desktop across Windows and Mac.
Step 2: Head over to the Copilot tab and click on it. The typical Copilot panel will pop out.
Step 3: Click the Tools button and select Agent Mode. Agent Mode is now activated.
Choosing your model: You’ll notice a model dropdown at the top. I’d highly recommend going with Claude Opus 4.5 from Anthropic. It’s their second newest model (Opus 4.6 is their current frontier model but isn’t yet available in Agent Mode). From my testing, Opus 4.5 is the best of the bunch compared to GPT 5.2 from OpenAI.
You’ll also notice a plus icon that lets you add content to the chat. You can attach files and add work docs, which gives Agent Mode a comprehensive context if you’re pulling in external resources alongside your workbook.
Building a dashboard from nothing
I wanted to build a client dashboard for a fictional consulting firm. I had no data to start with and only an empty workbook, but I did have Agent Mode.
Let’s go.
Prompt 1: Generate realistic data
I pasted the following into Agent Mode:
Create a worksheet called "ClientData" with 12 months of fictional consulting data for Greenfield Consulting. Include columns: Month, Revenue (£), Projects Completed, Avg Project Value (£), Client Satisfaction (%), New Clients, and Utilisation Rate (%). Make the data realistic with seasonal patterns and growth over the year.About 10 seconds later, the table appeared. Super quick. It built in realistic patterns too, with seasonal variation and key metrics all looking sensible across the 12 months.
Prompt 2: Build the dashboard
Now the fun part:
Create a new worksheet called "Dashboard" with three charts from the ClientData: a revenue trend line chart, a bar chart comparing Projects Completed vs New Clients by month, and a summary section at the top showing total revenue, average satisfaction, and average utilisation as headline numbers.Agent Mode planned it out step by step (you can see the reasoning in the chat pane), then created the Dashboard sheet with the summary statistics and all three charts.
Prompt 3: Refine the dashboard
The charts were functional but I wanted to tweak the charts styling:
Change the revenue chart colours to green and black only, and make the headline numbers larger and bolder.It updated the monthly revenue trend and bumped the summary numbers up to size 24. But here’s the thing, Agent Mode only updated one of the three charts. It missed the other two.
This is why the back-and-forth matters. You’re the human in the loop, the verifier and validator. I followed up asking it to make sure ALL charts follow the same colour scheme.
Both charts snapped into the green and black scheme immediately. Now all three charts were consistent, alongside the key summary statistics falling through from the client data.
Now, this is already useful. But there’s a problem.
The dashboard we built looks... fine. It uses the average Calibri font. We had to manually prompt it to change colours, and even then it missed two charts.
If we want consistency across future charts, we’d need to specify fonts, colours, and formatting rules every single time.
What if we could tell Agent Mode: “Every chart you build in this workbook should use my exact brand colours, my font, my formatting rules automatically”?
That’s what the “skill sheet” technique does. And it’s the single most underrated feature of Agent Mode that almost nobody is talking about.