Claude Cowork built a working spreadsheet

Most AIs can't build a spreadsheet you can actually open

June 01, 2026

In partnership with

Spreadsheets have a way of ruining a perfectly good afternoon. You ask an AI to build one, it hands back a file full of mystery numbers, and half the time you can't even open the thing. So you give up and rebuild it by hand anyway.

I read a LinkedIn post this week that put that exact frustration on trial, and I had to pass it along. The author ran an experiment most of us would never have the patience for. They asked 11 different AIs to build one single spreadsheet, and the results were rough.

By their account, Microsoft Copilot failed. Gemini failed. Grok failed. The detail that stuck with me: most of these tools couldn't even open the file they'd just created.

Then they tried Claude Cowork, and the difference was night and day. According to the post, Cowork built a working model with 6 tabs and over 700 formulas while the others tripped over the basics.

ChatGPT gives you generic answers because you give it generic prompts.

You know the fix: longer prompts, more context, clearer constraints. But typing all that takes five minutes per prompt, so you shortcut it. Every time.

Wispr Flow lets you speak your prompts instead of typing them. Talk through your thinking naturally — include context, constraints, examples — and get clean text ready to paste. No filler words. No cleanup.

Works inside ChatGPT, Claude, Cursor, Windsurf, and every other AI tool. System-level, so there's nothing to install per app. Tap and talk.

Millions of users worldwide. Teams at OpenAI, Vercel, and Clay use Flow daily. Free on Mac, Windows, and iPhone.

Why most people get AI spreadsheets wrong

The author named the trap most of us fall into, and I winced because I've lived it. We type something lazy like "make me a spreadsheet" and hit enter. The AI invents numbers out of thin air, hardcodes them, and buries them deep inside formulas you can't trace. When something looks off, you can't find the error. So you scrap it and start over by hand.

Their fix is less about a magic tool and more about a smarter process. That reframing is the whole reason I wanted to share this one.

The step-by-step playbook

Here's the exact sequence the post laid out. Each step earns its place, so I added a quick note on why it matters.

  1. Download the Claude app for free. No fancy setup, just the standard app.

  2. Click the "Cowork" tab. It sits between the Chat and Code tabs. That workspace is what makes real file-building possible.

  3. Select your folder, pick "Opus 4.8," and set "Effort high." The high effort setting is what buys you the deeper, more careful build instead of a rushed one.

  4. Connect Google Drive using the "+" button under Connectors. This is the trick that lets the finished sheet open live, instead of vanishing into a download you can't locate.

  5. Paste your data and describe the sheets you need. Be specific about the tabs and the structure you want.

  6. End every prompt with one specific line. This is the part that actually changes the outcome.

That last step is the heart of the method. The author says to close every prompt with this exact instruction:

"Before building, list your top 10 assumptions so I can sanity-check them, then execute."

They called that line the whole trick, and I agree. It flips the power dynamic. Instead of the model running off and guessing, it pauses and shows its work first. You steer it, not the other way around.

“Who is this person again?”

You’ve had that moment. Walking into a call, scrolling through old emails, trying to remember what you promised. Lindy texts you a brief 15 minutes before: attendee context, past discussions, open items, talking points. All pulled automatically. Try Lindy free.

*Ad

Three things the author didn't expect

What I liked about the post is that the writer was honest about being surprised. Three things caught them off guard.

  1. It defends its numbers before it builds. The AI lists its 10 assumptions and you get to challenge them. Every input lives on a dedicated "Assumptions" tab, fully editable. Nothing gets buried inside a formula you can't follow.

  2. It builds board-ready models from scratch. They handed it a prompt like "12-month revenue forecast, 4 service lines, Base/Bull/Bear toggle" and it built the whole thing. Funnel, P&L, Dashboard, Scenarios. It didn't waste time explaining what a forecast is. It built one.

  3. It opens straight into Google Sheets. No more "done!" message stapled to a file you can never find again. You click Drive and the full model opens, live and working.

Why this matters beyond spreadsheets

The line from the post that really stuck with me was about expertise. The author said 41 years of Excel experience is no longer a moat. That's a bold claim, but their test backs it up. Claude built in roughly 7 minutes what used to eat a full day of manual work.

The bigger lesson runs past spreadsheets, though. The real shift is in how we talk to these tools. Asking for assumptions first isn't just an Excel hack. You can bolt that same sanity-check line onto almost any complex request, whether you're building a budget, drafting a project plan, or mapping a content calendar. It forces the model to be transparent before it commits.

It fits a pattern I keep noticing. The people getting the most out of AI aren't the ones with the slickest one-line prompts. They're the ones who build a little verification into the process and refuse to take the first output at face value.

My honest take

I've thrown away more AI-built spreadsheets than I'd like to admit, always for the same reason: I couldn't trust a single number in them. So the fix here lands for me, and it has nothing to do with Cowork being magic. It's that one assumptions line doing the heavy lifting.

That's the part worth stealing even if you never touch a spreadsheet this month. Make the model show its assumptions before it runs, and you turn a black box into something you can actually argue with.

Tonight, take whatever you were about to ask an AI to build, and add this to the end of your prompt: "Before building, list your top 10 assumptions so I can sanity-check them, then execute." Read the list before you let it run. See how much cleaner the result comes out.

How Jennifer Aniston’s LolaVie brand grew sales 40% with CTV ads

The DTC beauty category is crowded. To break through, Jennifer Aniston’s brand LolaVie, worked with Roku Ads Manager to easily set up, test, and optimize CTV ad creatives. The campaign helped drive a big lift in sales and customer growth, helping LolaVie break through in the crowded beauty category.

*Ad