Thursday Thoughts: Claude Code Is Lotus 1-2-3, and Copilot Is Playing the Excel Game

I’ve been reading up on the history of Excel this week, mostly for fun, and I stumbled into an argument I wasn’t expecting to make. It starts with a claim that sounds like a hot take and turns out to be closer to a well-documented fact: enterprise knowledge workers have been vibe coding for decades. It was just called VBA.

That’s not a cute reframe. “End-user computing” is an actual, decades-old risk-management category — EUC applications facilitate the production of working applications by non-coders, and can essentially be thought of as a subset of shadow IT. Finance teams, ops teams, sales teams have been building unsanctioned little pieces of software inside spreadsheets since the 1980s, without a CS degree or a ticket in the backlog. Coding agents didn’t invent this. They just made it faster and put a chat window on it.

But the more I dug into Excel’s history, the more I realized the interesting story isn’t the vibe-coding parallel. It’s what happened to the other spreadsheet — the one that was actually winning.

The Best Product Lost the 1980s

Excel launched in 1985, Mac-only. Lotus 1-2-3 launched in 1982 and by 1988 — three years after Excel existed — still held roughly 70% of the spreadsheet market to Excel’s 10%. Lotus wasn’t just ahead. It was the category. Wall Street had already fallen in love with spreadsheets years before Excel showed up.

Then Lotus made a bet: it stayed focused on OS/2, the IBM-Microsoft joint venture operating system, instead of porting 1-2-3 to Windows. Microsoft, meanwhile, switched to promoting Windows 3.0, which became the dominant operating system in the world, leaving Lotus’s products stuck on an old character-based interface. One Lotus insider, years later, called it exactly what it was: “It was a head fake by Microsoft.” By the early 1990s, Excel had started to outsell Lotus 1-2-3, and Lotus never recovered — IBM bought what was left of it in 1995.

Read that sequence again. It’s not “the better product won.” Depending who you ask, it might genuinely be the opposite: some accounts argue Excel earned it on merit (Excel was a vastly superior Windows experience and the first “killer app” for that environment), and others just point at the calendar (Lotus was surpassed by Microsoft in the early 1990s largely because it didn’t take Windows seriously in time). Either way, the deciding factor wasn’t a feature checklist. It was which company controlled the platform the entire market was about to move onto.

I think we’re watching a rerun.

Claude Code Is Lotus. Copilot Is Betting It’s Excel.

Every satisfaction survey right now says the same thing: Claude Code is the better product. Among developers with 10+ years of experience, 46% now choose Claude Code versus just 9% who prefer Copilot, and that gap holds up across independent surveys — Claude Code leads on “most loved” at 46%, against Copilot’s 9%.

And yet: GitHub Copilot controls enterprise deployment at 90% of Fortune 100 companies, and at companies with 10,000+ employees, Copilot leads at 56% adoption, reflecting enterprise procurement inertia, existing Microsoft 365 integrations, and compliance infrastructure. Meanwhile Claude Code dominates the small end of the market, where there’s no procurement inertia to fight through, holding 75% adoption at companies under 50 employees.

That is, functionally, 1988 again. The better product is winning where switching is easy and losing where a platform relationship already exists. If I stopped the post here, this would practically write itself: Claude Code is Lotus, Copilot is Excel, distribution beats features, wait a few years for the enterprise numbers to flip the way they did in 1992.

I don’t think that’s the actual story, though. I think it’s more interesting, and slightly worse for Anthropic, than that.

This Time, Microsoft Isn’t Waiting Around

Lotus’s fatal mistake wasn’t losing a feature war. It was assuming distribution would keep working while the product quietly fell behind. Microsoft, of all companies, knows this playbook better than anyone alive, because it ran it. And what’s actually happening inside Copilot right now doesn’t look like a company hoping its product improves eventually. It looks like a company that skipped straight to the endgame.

Start with the product gap itself: Copilot Cowork, Microsoft’s new long-running agentic product, is built on Anthropic. Microsoft doesn’t need Copilot’s underlying model to beat Claude. It can just license Claude, wrap it in Microsoft’s distribution, and call it Copilot. That’s a move Lotus never had available to it, and it quietly dissolves the entire “better model wins eventually” framing. If the engine inside both products increasingly comes from the same place, the competition was never really about the model.

Then there’s the part that would have saved Lotus if anyone at Lotus had thought of it: governance, shipped on day one instead of bolted on 30 years later. It took the Excel-industrial-complex three decades to build EUC risk-management platforms because spreadsheets went ungoverned for so long that regulators had to force the issue — Shadow IT and uncontrolled end-user computing tools represent one of the most significant and underestimated operational risks in financial services, and regulators are increasingly requiring inventories, risk assessments, and controls. Microsoft is not waiting three decades this time. Agent 365 is already shipping as a $15-per-user control plane for IT and security teams to observe, manage, and secure agents, bundled into a new Microsoft 365 “Frontier Suite” that wraps Copilot, governance, and security tooling into one CIO-friendly line item — a direct response to the fact that 86% of IT leaders say they need additional governance to manage agents at all.

That’s the tell. Microsoft isn’t trying to out-model Anthropic. It’s trying to become the thing every EUC governance vendor spent 30 years trying to retrofit onto Excel, except this time it’s native, and it’s Microsoft selling it to the same CIOs who already buy everything else from Microsoft.

The Counterargument: Anthropic Isn’t Lotus, Either

Let me argue against myself, because the analogy is too clean if I don’t. Lotus’s failure was passivity. It sat on distribution and assumed the product didn’t need to keep up. Anthropic is doing the opposite of that: Anthropic launched Cowork to bring Claude Code’s capabilities to knowledge workers beyond developers — financial analysis, legal review, sales operations — which is exactly the move Lotus never made. Lotus never tried to become the platform. It tried to stay the best spreadsheet. Anthropic is explicitly trying not to make that mistake.

There’s also a real internal signal worth sitting with: Microsoft internally adopted Claude Code across major engineering teams for complex work, notable given that Microsoft sells GitHub Copilot. If Microsoft’s own engineers reach for the competitor’s product when the work actually matters, that’s not nothing. It’s the modern version of the diehard Lotus fan who refused to switch, except it’s happening inside the building that owns the incumbent.

So maybe the honest framing isn’t “Copilot wins” or “Claude Code wins.” It’s that the thing worth owning was never the model. It’s the governance layer standing between “agent” and “agent your compliance team will actually sign off on.” Whoever wins that layer wins the enterprise, the same way whoever won the OS won the spreadsheet. Right now Microsoft is moving faster on that specific bet than Anthropic is, and it’s the one lesson from the Lotus story that Microsoft, uniquely, has no excuse not to have learned.

Hold Me to It

Here’s a marker I’m comfortable being wrong about: by the end of 2027, enterprise agent procurement decisions will be driven more by governance and control-plane attach rate — audit trails, permission scopes, IT sign-off — than by which model scores higher on a coding benchmark. If Claude somehow wins that fight instead of just the model fight, I’ll write the follow-up admitting Anthropic solved the one thing Lotus never even tried to.

Also worth remembering: the market-share numbers themselves don’t agree with each other. Depending which 2026 survey you read, Claude Code’s overall share is somewhere between 18% and a majority of the market. That spread alone tells you this fight isn’t settled. Neither is the one that mattered thirty-some years ago before the first Windows PC showed up on someone’s desk with Excel already on it.

If you’re inside an enterprise right now, which one did IT actually let you install?

By the Numbers

• 70% vs. 10% — Lotus 1-2-3’s spreadsheet market share against Excel’s in 1988, three years after Excel launched
• 1992 — the year Excel decisively overtook Lotus 1-2-3, roughly seven years after its 1985 launch
• 46% vs. 9% — developer preference for Claude Code over Copilot among engineers with 10+ years of experience
• 90% — the share of Fortune 100 companies where GitHub Copilot controls enterprise deployment today
• 75% — Claude Code’s adoption rate at companies under 50 employees, where there’s no procurement inertia to overcome
• $15/user — the price of Microsoft’s Agent 365 governance control plane, shipped years into the agentic era, not decades
• 86% — the share of IT leaders who say they need additional governance just to manage the agents already in their organizations
• 0 — the number of platform shifts Lotus survived by assuming distribution alone would keep it in first place