The spreadsheet is not going anywhere

Spreadsheets have survived every productivity revolution of the past forty years. Databases, cloud software, dashboards, and automation tools have all promised to replace them, and yet Excel remains the shared language of business. Every industry uses it. Every skill level touches it.

What has changed is the kind of help available when you are stuck, slow, or trying to push your work further. AI agents can now read your actual spreadsheet files, understand the formulas inside them, and either fix problems or build things from scratch. A financial analyst recently described using one to build a complex multi-sheet model with circular references in a single day, work that would normally take a full week. That is the kind of shift this page is about.

This guide is organized by skill level. Jump to the section that matches where you are today.

What is an AI agent?

If at this point you are asking yourself what exactly an AI agent is and how it is different from ChatGPT, you are not alone. A chatbot is a conversation in a browser window. You type, it responds, and everything stays inside that chat. An AI agent goes further because it connects to your actual tools, like your files, browser, or calendar, and takes action on your behalf instead of just generating text.

In this guide, the agent reads your Excel files directly, understands the formulas and data inside them, and creates or modifies spreadsheets without you copying and pasting anything. To understand the full difference between chatbots and agents, read our breakdown of AI agents vs. chatbots.

What you need before starting

This guide uses Claude Cowork. If you have not set it up yet, start with our setup guide. For Excel work specifically, here is what to know:

  • Your Excel files in an accessible folder. When you start a session, grant Cowork access to the folder where your spreadsheets live. It reads and creates .xlsx, .csv, and .xlsm files natively. Excel does not need to be open or even installed on your machine for Cowork to work with your files.
  • Clear file and sheet names. If you are working with multiple files or sheets, name them clearly before your session. A file called "Q1_Revenue_Report.xlsx" will produce more precise results than "Book2.xlsx."
  • No formula knowledge required. You describe what you want the spreadsheet to do. The agent figures out the formulas. You do not need to know Excel syntax to get real value from this.

For beginners: less confusion, more confidence

If your relationship with Excel is mostly storing data in rows and columns, with the occasional SUM formula, this section is for you. A lot of people feel limited by what they know how to do. These use cases remove that ceiling.

Clean and organize messy data

You downloaded a report from your accounting system, or received a spreadsheet from a colleague, and it is a mess. Numbers stored as text, inconsistent date formats, extra spaces in every cell, blank rows scattered throughout. Cleaning this manually is tedious and easy to get wrong.

"I have a spreadsheet called 'Customer_Data.xlsx' in this folder. The data is messy and I need it cleaned up before I can use it. Specifically: remove all duplicate rows, trim extra spaces from the text in every column, make sure all dates in the 'Date' column are formatted consistently as MM/DD/YYYY, convert any cells where numbers are stored as text into actual numbers, and remove any completely blank rows. Save a cleaned version called 'Customer_Data_Cleaned.xlsx' and give me a summary of everything you changed."

The summary at the end matters. It tells you exactly what was fixed and how many rows were affected, so you are not left guessing about what changed.

Build a simple budget or expense tracker

If you want a personal or team budget tracker but do not know where to start, describe what you need and let the agent build it.

"Please create a monthly budget tracker spreadsheet for me. I want to track my income and my expenses by category. My categories are: rent, groceries, utilities, transport, subscriptions, dining out, and miscellaneous. I want to see how much I have budgeted for each category, how much I have actually spent, and the difference between them. Include a totals row showing total income, total budgeted expenses, total actual expenses, and how much I have left over. Highlight any categories where I have gone over budget in red. Save it as 'Budget_Tracker.xlsx'."

You get a working spreadsheet with all the formulas already in place. All you do is fill in your numbers.

Understand what a formula is doing

You have inherited a spreadsheet with a formula you do not understand. You can paste it directly into the chat and ask for a plain-English explanation.

"Can you explain what this formula does in plain English, and tell me if there is a simpler way to write it? =IF(AND(B2>0,C2<>""),IFERROR(VLOOKUP(A2,Sheet2!$A:$D,4,FALSE),"Not Found"),"") I use this in a sales tracker but I am not sure what happens when the result is empty."

This is one of the fastest ways to stop feeling intimidated by formulas you did not write yourself. Understanding what something does makes you able to modify it confidently.

Turn raw data into a readable summary

You have a spreadsheet full of data and need to present it to someone who does not want to look at a wall of numbers. The agent can create a clean summary sheet ready to share.

"I have a sales data file called 'Sales_2025.xlsx'. It has monthly figures for each of our five products across twelve months. Create a new sheet in the same file called 'Summary'. Show the total sales per product for the full year, the best month for each product, the worst month for each product, and the overall monthly total across all products. Make the table look clean and professional. Then add a bar chart showing total annual sales by product and place it below the table."

For intermediate users: more power, less manual work

If you know your way around Excel and can build things yourself, the bottleneck is usually not skill, it is time. You spend too much of it on repetitive tasks, formatting work, and debugging formulas that should not take that long. This section covers where an agent saves the most time at your level.

Build a dashboard from existing data

Dashboards take time to build well, not because any individual part is hard, but because there are a lot of moving pieces. Setting up the source data correctly, writing aggregation formulas, creating charts, and making sure everything updates dynamically all adds up to a half-day project minimum.

"I have a sales report in 'Monthly_Sales.xlsx' with data across four sheets, one per region: North, South, East, and West. Each sheet has the same columns: Date, Salesperson, Product, Units Sold, and Revenue. Create a new 'Dashboard' sheet that pulls together the key metrics from all four regions. I want: total revenue by region for the current year, the top three salespeople by revenue across all regions, a month-over-month revenue trend for the overall business, and a product breakdown by units sold. Add charts where they make sense. Build it so the dashboard updates automatically when I add new data to the regional sheets."

Specifying that you want dynamic formulas rather than hardcoded values is worth adding to any dashboard prompt. The agent defaults to building things that update automatically, but being explicit removes any ambiguity.

Create a template you can reuse every month

If you recreate the same report every month by copying last month's file and editing it, that is a workflow worth fixing once.

"I want to create a reusable monthly reporting template for our team. Each month we report on: leads generated, leads converted to opportunities, opportunities closed as won, opportunities closed as lost, and total revenue. We track this by salesperson and we have eight people on the team. The template should have an input section for the month's data, a section that automatically calculates team totals and individual conversion rates, and a chart section visualizing the key numbers. Build it so I can enter a new month's data without touching any of the formulas. Save it as 'Monthly_Report_Template.xlsx'."

Write and explain the formulas you avoid

There is a category of Excel formulas most intermediate users know exists but tend to avoid: INDEX/MATCH, XLOOKUP, dynamic array functions like FILTER and UNIQUE, and anything involving SUMIFS with multiple conditions. Ask the agent to write them and explain them, and you will understand the logic well enough to adapt them yourself.

"I have two sheets in a workbook. Sheet 1 has customer orders with columns: Order ID, Customer Name, Product, Quantity, and Price. Sheet 2 is a product catalog with columns: Product Name, Category, and Cost Price. I need two formulas in Sheet 1. The first should look up each product in Sheet 2 and return its category. The second should calculate the profit margin for each order, which is the price minus the cost from Sheet 2, divided by the price. Write both formulas and explain how each one works so I can adapt them if the columns change."

Audit and fix broken formulas

Nothing wastes time like a spreadsheet where one formula is broken and the error has cascaded to five other cells. An agent can trace the source.

"My spreadsheet 'Budget_Model.xlsx' has a series of #REF! errors appearing in the Summary sheet. I think the issue started when I deleted some columns from the Raw Data sheet last week, but I cannot trace it. Open the file, find where the errors originated, fix them, and explain what caused the problem. If there are other broken references that are not showing as errors yet, flag those too."

For advanced users: work that used to take days

A financial analyst recently posted about using an AI agent on a complex model with circular references and logic spread across ten sheets, the kind where one wrong cell ruins everything. Their conclusion: a model that normally takes five days took one day. The prompts in this section are built for users working at that level.

Audit a complex financial model

Complex models accumulate problems over time. Formulas that work but are fragile, dependencies that are hard to trace, assumptions buried in cells with no documentation, and errors that only surface under specific conditions. Getting an outside audit, even from an AI, catches things that familiarity hides.

"I have a financial model in 'DCF_Model.xlsx' that I need audited before presenting to investors. The model has ten sheets and includes circular references in the debt schedule. Please: trace all formula dependencies across sheets and flag any that appear fragile or incorrect, identify any hardcoded values that should be formula-driven, verify that the circular references resolve correctly, check for any cells where the formula logic does not match the label or adjacent cells, and summarize the five highest-risk areas in the model. Document your findings in a new 'Audit Notes' sheet inside the same file."

Build a financial model from a brief

Describe the structure you need and have the agent build the scaffolding, rather than starting from a blank sheet yourself.

"I need a three-statement financial model for a SaaS business with an income statement, balance sheet, and cash flow statement. Key assumptions: annual recurring revenue starts at $2M and grows 40% per year for five years, gross margin is 72%, operating expenses are split between sales and marketing at 35% of revenue, R&D at 20%, and G&A at 10%. Build a separate Assumptions tab where I can change the key inputs and have everything update automatically. Add a DCF valuation tab using a 12% discount rate and 3% terminal growth rate. Set up the structure and formulas correctly. I will fill in the historical actuals afterward."

Build a scenario analysis and sensitivity table

Scenario analysis is powerful but time-consuming to build properly. Getting the layout right, linking the scenarios to a single assumptions driver, and formatting the sensitivity table clearly can take most of a day.

"I have a pricing model in 'Pricing_Model.xlsx'. I want to add a scenario analysis section with three cases: base case using current assumptions, optimistic case with 20% more units sold and a 5% price increase, and conservative case with 10% fewer units sold and a 3% price decrease. Build a scenario summary table showing all three side by side with revenue and gross profit. Then build a two-variable sensitivity table showing how net profit changes as unit volume varies from negative 30% to positive 30% in 10% steps, and average selling price varies from negative 20% to positive 20% in 5% steps. Color the sensitivity table from red to green so the best outcomes are immediately obvious."

Write and document a VBA macro

If you repeat the same sequence of Excel actions every week, a macro can handle it. Most users avoid writing VBA from scratch because the syntax is tedious and the debugging is painful. Describe what you do manually and ask the agent to write it.

"Every Monday I do the following in Excel manually: open the sales report file, copy the previous week's data from the Raw Data sheet into a new row in the Weekly Archive sheet, update the date in cell A1 of the Dashboard sheet to reflect the current week, and save the file. I want a VBA macro that does all of this automatically. Write the macro code, explain what each section does in plain language, and tell me step by step how to install it in the workbook."

Getting the most out of your agent for Excel work

Name your files and sheets clearly before starting. The more precise your file structure, the more precisely the agent can work. Vague names produce vague results.

Ask for documentation inside complex models. When having the agent write formulas or build model logic, ask it to add cell comments explaining what each formula does. This pays off every time you come back to the file later.

Request a change summary. After any session where the agent modifies an existing file, ask it to list exactly what changed. This makes the review process straightforward and gives you a record of what was done.

Keep a backup. Before giving an agent access to an important file, copy it to a backup folder. It takes ten seconds and removes all the risk from the session.

Iterate without frustration. If the first result is close but not quite right, describe what needs to change. Complex spreadsheet work rarely lands perfectly in one pass, and that is normal. Each follow-up prompt gets you closer.