Excel Is Dead. Long Live Excel.

"Excel is dead." It has been said many times, in one form or another, by many people selling some form of SaaS. Even Figma's CFO, Praveer Melwani, when asked on the a16z Show whether Excel is under threat, said: "No, it's like my love language."

These aren't normal times, so I'll say it now. Excel is on life support, and it's going away.

In February I wrote that what happened to software is happening to finance. I wrote it because that was when AI eclipsed me at writing code: it had become easier to tell it what to write than to write it myself. Excel is at that point now.

What Excel actually does

Strip it down and Excel does three jobs.

1. It presents financial concepts in a format humans can digest.
2. It's a calculation engine.
3. It's a database, for those of us who misuse it as one.

The first job isn't going away. Excel does the other two horribly, and with horrible version control: a file name ending in _v7_final_FINAL, and two versions compared by opening them side by side and scrolling. A sheet still tops out at about a million rows.

I learned the database lesson the hard way. Early in my career I demanded the biggest Xeon processor I could get so I could run 100,000-row workbooks stacked with INDIRECT formulas. I cringe at that now. I was using Excel because I didn't understand SQL, or how to move the work into the tools built for it.

Built for humans

At its heart Excel is nearly 50 years old. VisiCalc shipped in 1979, Lotus 1-2-3 in 1983, Excel in 1985, and the grid hasn't changed much since. It's built for humans and around human limits. We can't hold more than two dimensions in our heads at once, so everything goes on a flat grid. We need to see every intermediate step, so every number gets its own cell. Those are reasonable choices for a person doing arithmetic by hand. They are strange choices for a machine.

We're about to stop needing a calculation tool built for humans, the same way we'll soon stop needing steering wheels in cars. There are better technologies for the engine: pick your programming language, put the data in Postgres, and use HTML, JavaScript and CSS for the part a person looks at.

What won't change is the tabular presentation of information, so our weak human brains can weigh in on things. The engine behind it doesn't matter. Do any of us really care whether a number came from a VLOOKUP, a SELECT * FROM, or assembly? I don't, as long as I can see the result and trust where it came from.

I'm confident a superior alien species, handed the job of showing a person a financial problem, would not invent Excel.

It already happened to PowerPoint

A year ago it was almost unheard of to present information in anything but PowerPoint, Google Slides or a PDF. Now, at the AI-forward companies I work with, a Claude artifact is a normal thing to hand someone: a page that renders the numbers and the argument together, and that the model can change as easily as a person can.

Spreadsheets are next. For several months I've been working in my own spreadsheet artifact, built to be really easy for LLMs to work with, with git underneath so every change has a history and any two versions can be compared. This week Anthropic launched Claude Sheets, a spreadsheet a team edits together with Claude, which works in the cells alongside them and lays out each change it made.

That's the future. The grid stays, because it's how we take a problem in. The work in the cells moves to the machine.

But then a curious thing happened: I stopped using spreadsheets and artifacts almost entirely. The models and the harnesses got so good that I work at a much higher level now. I'm focused on why we build spreadsheets, which is to solve problems, rather than on the art of building them.

I'm a spreadsheet artist at heart

This saddens me on some level. When I was an investment banker I prided myself on how good I was at Excel. We used to hold model offs: a room full of type-A analysts, 1am, everyone racing to build the same model fastest and cleanest. It was a way to believe we weren't wasting our lives for less than minimum wage on the promise of a high-paying job later. I don't touch the mouse. To this day I won't use a keyboard without a 10-key and a proper page up and page down cluster, because my hands know where everything is. If you've never seen that world, KRAZAM's MAKRO, a parody of a pro gamer streaming Microsoft Excel, is closer to the truth than it should be. It was so accurate that eventually life imitated art, and Excel became an esport.

I'll still take any of you in a model off. But nobody is paying for my keystrokes anymore. They want a business problem solved.

At the same time, I now build and refactor far more complicated models than I ever did in a grid, with real calculation engines, and I can plug in any open-source technology on the planet. None of it was ever about the model. I've never walked into a boardroom and bragged about the model I built to get the answer. The point is getting an operating decision in front of my eyes as fast as possible.

What's still hard

Three things still limit this, and they're where I spend my time.

Connecting data, so nothing gets pasted in by hand. Version control, so every number has a history; mine runs on git. And permissions. In finance we hold sensitive data, and we have to control exactly which humans and which LLMs can see and change which numbers.

The rest is for the bots.