BuildData,NotSoftware

TheBetEveryoneIsGettingWrong

The prevailing wisdom in Silicon Valley right now: AI will turn everyone into a programmer. Andrew Chen, general partner at a16z, put it plainly: "AI code gen means that anything that is currently modeled as a spreadsheet is better modeled in code. You get all the advantages of software — libraries, open source, AI, all the complexity and expressiveness."1

Here's what he's getting right: AI does replace software. Custom apps, internal tools, dashboards — AI collapses all of that. But here's what he's getting wrong about what's left.

What's left isn't code. It's data. When AI strips away the software layer, people don't become programmers. They become data processors — structuring, modeling, and reasoning about information. And the tool for that job has been here for 47 years.

ADayintheLife

It's 8:47 AM. Sarah is a financial analyst at a mid-cap industrial company. She opens Outlook — 34 unread emails, three meeting invites, a Slack thread about Q2 numbers that's already 40 messages deep. She spends the next 90 minutes triaging: responding, forwarding, declining a meeting that could've been an email.

At 10:15 she finally opens the file that matters: a spreadsheet. It's the Q2 forecast model. She adjusts three assumptions, watches the downstream numbers cascade, checks the sensitivity table, and screenshots the summary for her VP. This takes 25 minutes — and it's the most valuable thing she'll do all day.

By 10:40 she's back in meetings. Stand-up, then a "quick sync" that runs 45 minutes. Lunch. Another meeting. At 3 PM she gets 40 minutes to update the board deck — pulling numbers from the same spreadsheet. At 4 PM, Salesforce needs her pipeline data entered. She copies it from a spreadsheet she's been maintaining on the side because Salesforce's forecasting is useless for her purposes.

Sarah's day is 480 minutes. About 275 of those go to communication — email, chat, meetings, searching for context. The rest splits across documents, domain apps, and spreadsheets. But the 60 minutes she spends in Excel? That's where the actual thinking happens.

These aren't guesses. Microsoft's 2025 Work Trend Index — 31,000 workers across 31 countries — found 57% of the workday goes to communication.2 Hubstaff's 2026 report — 140,000 workers, 17,000 organizations — found the average worker gets just 2 to 3 hours of focus time per day.3 Acuity Training surveyed 1,000 office workers and found 38% of their work involves a spreadsheet. Finance professionals average 2.5 hours a day in Excel.4

But the numbers miss the point. It's not about how much time goes where — it's about what each minute is worth. Sarah's 275 minutes of communication keep the machine running, but they don't create value. Her 60 minutes in a spreadsheet produce the analysis that drives a $400M division's capital allocation. Minute for minute, the spreadsheet work is where the leverage lives.

ThenSheGoesHome

At 6 PM, Sarah closes her work laptop. She opens her personal laptop and opens... a spreadsheet.

The household budget. The vacation comparison — Lisbon vs. Barcelona, flights and hotels side by side, a formula that calculates total cost per person including exchange rates. Her daughter's college savings projection. The remodel contractor bids — three quotes, line-item comparison, a column she added for "gut feeling score 1-10."

Nobody told her to use a spreadsheet for any of this. There's Mint for budgeting, TripIt for travel, a dozen apps for contractor management. She's tried some of them. She always comes back to the grid.

Why? Because she can see everything at once. Because she can add a column whenever she wants. Because she doesn't have to learn someone else's system for organizing her own life.

AIReplacesSoftware.NotData.

Let's give Andrew Chen his due. He's right that AI replaces software. The custom internal tool, the dashboard nobody maintains, the Retool app that took two sprints to build — AI collapses all of that. You describe what you want in plain English and the software appears.

But what's underneath the software? Data. Rows. Columns. Assumptions. Models. The software was always just a view on top of data. When AI strips the software layer away, what remains is the data layer — and the human who understands what the data means.

Chen's mistake is thinking that when AI makes software trivial, people will build more software. They won't. They'll skip the software entirely and go straight to the data. They already do. That's what a spreadsheet is.

TheExpandingMiddle

Here's the sequence:

First, AI kills the easy software. The dashboards, the CRUD apps, the internal tools. People who used to request those from engineering will just... not. They'll ask AI for the answer directly, or they'll model it themselves in a spreadsheet with AI sitting next to them.

Then, AI kills the hard software. The BI platform, the reporting layer, the analytics pipeline. Why maintain a data warehouse when you can ask Claude to query the source data directly?

What's left is a massive, expanding middle: people working directly with data, using the simplest possible tool — the grid. AI handles the plumbing. Humans handle the judgment. The spreadsheet is the interface between the two.

Code is a black box. Spreadsheets are glass. When the CFO asks "where did this number come from?" — you point to cell D47. Try that with a Python script Claude generated.

TheBestSpreadsheetThatEverExisted

Here's what the "spreadsheets are dead" crowd misses entirely: AI doesn't threaten the spreadsheet. It removes every limitation the spreadsheet ever had.

Three things have always held spreadsheets back. First, the formula ceiling — most users know SUM, AVERAGE, maybe VLOOKUP. The real power (INDEX/MATCH, array formulas, dynamic ranges, LAMBDA) lives behind a learning curve that 90% of users never climb. Second, the data janitor problem — half of spreadsheet time isn't analysis, it's cleaning. Reformatting dates, splitting names, deduplicating rows, normalizing categories. Third, the scale wall — past a few thousand rows, spreadsheets get slow, formulas get brittle, and you start hearing "you should really use a database for this."

AI attacks all three. And not with some new tool you have to learn. With formulas.

ItLivesintheFormulaBar

In August 2025, Microsoft shipped the =COPILOT() function.6 It lives in the Excel formula bar — right next to SUM, VLOOKUP, and IF. You type =COPILOT("categorize this customer feedback as positive, negative, or neutral", A2) and it returns the answer in the cell. It references ranges, auto-updates when data changes, and chains with other formulas. AI became a spreadsheet function.

Google matched it in March 2026 with "Fill with Gemini" — highlight a range, describe what you want, and Gemini populates every cell. Google claims 9x faster than manual entry. Their Sheets AI hit 70.48% on the SpreadsheetBench dataset — nearing human expert performance.8

And in Google Sheets, you can type =CLAUDE("summarize this vendor proposal", A2:F2) — Anthropic ships an official add-on that turns Claude into a cell-level function.9 Run it across an entire column. Classify 10,000 rows of customer feedback. Extract key terms from contracts. All in the grid.

These tools are early. =COPILOT() is still in beta. Gemini's benchmark means it fails 30% of the time. But the trajectory is unmistakable — and every product ships inside the grid, not outside it. The AI came to you.

Sarah'sSpreadsheet,March2026

Same Sarah. Same Q2 forecast model. But now she has Copilot in Excel — and in December 2025, Microsoft shipped Agent Mode:7 an AI that plans, executes, and validates multi-step tasks right in the grid.

She doesn't write the XLOOKUP that cross-references vendor pricing from three tabs. She describes what she wants: "Pull the latest unit cost for each SKU from the vendor quotes tab, fall back to last quarter if no current quote." The formula appears. It's correct. It would have taken her 20 minutes to write and debug. It took four seconds.

The messy CSV from procurement? She used to spend 45 minutes cleaning it — dates in three different formats, company names with trailing spaces, duplicate rows with slightly different capitalization. Now Agent Mode handles the whole pipeline: import, reshape, deduplicate, format. She reviews the result, not the process.

At home, she opens her vacation comparison in Google Sheets and types =CLAUDE("which destination is the better value considering weather in June, flight time from Dallas, and total cost for a family of four?", A1:G10). The answer appears in a cell. She can see the data it used. She can change an assumption and ask again.

But here's the part that matters most: she's still in the grid. She can see every formula. She can trace every reference. She can click on any cell and understand exactly what it's doing and why. The AI didn't take her out of the spreadsheet — it took the friction out of the spreadsheet.

She used to spend 60 minutes in Excel and wish she had more time. Now she spends 60 minutes and gets three times as far.

This is why the spreadsheet doesn't just survive AI — it enters its golden age. For 47 years, the spreadsheet's power was gated by your willingness to learn its language. VLOOKUP syntax. Nested IF statements. Pivot table configuration. The people who climbed that curve got enormous leverage. Everyone else got a fraction of what the tool could do.

AI removes the gate. =COPILOT() in Excel. =CLAUDE() in Google Sheets. Agent Mode that builds your model while you describe it. Gemini that solves optimization problems you didn't even know had a name.

Microsoft reports that Copilot users save 40 to 60 minutes per day.10 Google says 73% of Gemini enterprise accounts use it in Workspace.11 These aren't toy demos. They're production tools being used by millions of people — inside the same grid they've used for decades.

The addressable market for "spreadsheet power user" just went from 50 million to 1.5 billion. Every single user becomes the expert they never had time to become. And they didn't have to learn a single new tool to get there.

AndThen?

Eventually, AI takes the data processing too. The models get good enough to not just query data but to understand context, make judgment calls, and recommend decisions without a human structuring the inputs.

But that's not 2026. That's not 2028. The gap between "AI can write code" and "AI can replace human judgment about what a business should do" is enormous. In that gap — which could last a decade — the spreadsheet doesn't just survive. It thrives. Because AI made it accessible to everyone and powerful for anyone.

Sarah doesn't need a better app. She needs her spreadsheet to be smarter. AI gives her that. The remodel contractor bids, the Q2 forecast model, the vacation comparison — none of these need software. They need a grid, a formula engine, and an AI copilot that turns "what I want" into "what I see."

TheThreeErasoftheSpreadsheet

StressTest

Every thesis has blind spots. Three reviewers stress-test this one.