How Windows got to version 3 – an illustrated history

Opinion Windows 1 and 2 flopped almost as badly as OS/2 did. How did Microsoft stage one of the greatest comebacks ever with Windows 3?

Earlier this month, we took a look at how Microsoft learned important lessons from the failure of OS/2, even though less than a decade later, it had already started forgetting what went wrong and why. Such organizational amnesia has a long history. For instance, in the middle of the 1700s, the Royal Navy worked out how to cure scurvy, but by the end of the 1800s, it had forgotten again.

Our OS/2 article was actually written just before 2024's holiday season. So, for this itinerant historian of operating systems, New Year's Day 2025 held an unexpected treat: Nina Kalinina's Windows 2: Final Fantasy of operating systems. We know that some of our stories go on a bit too long, but this one outdoes us at well over 10,000 words. We recommend it highly, for all that. You'll have to read carefully and closely. These pointers are not always obvious.

This history is still important. To make sense of the industry today, you need to know where things came from and how they evolved. Some big products with huge investment flopped; some trivial niche products ended up dominating whole industries. When you have history, with citations, you can demolish many conspiracy theories – some of which are still around and were aired in the comments to our previous article. Knowing how some of this pivots on just two or three people pulling a few all-nighters really explodes some of the corporate blather that too many people still take as gospel today.

The article isn't very linear, so we thought we'd point out some of the highlights and some of the obscure details, and then spell out what we feel are the important conclusions to draw from it. Kalinina anatomizes the first three Windows versions, the ones that had to be installed on top of MS-DOS, but the structure is a little disorienting.

It starts with a guided tour of Windows 2, showing some primitive, clunky tools that are recognizably the ancestors of programs that run a million businesses around the world today. Then, unexpectedly, the story jumps backwards in time, all the way to the late 1960s, to explain the historical context in which Windows was first designed and how it gradually developed.

This second section is much longer, and it's a chronology. It looks at the development of all the major GUI OSes, so although it focuses on the most significant versions of Windows through the 1980s, it also compares and contrasts it with its contemporaries: classic MacOS, Digital Research GEM, Commodore AmigaOS, OS/2, and others. Examining the changes from Windows 1, to 2, to 2.1, to 3, almost uniquely, it explains why things changed, and what was happening in the industry, especially around GUIs, to permit or cause them to change. On the way, it demolishes a few myths about Windows/386 and Windows/286. By showing what the different offerings looked like at the same time, it shows Windows gradually changing from an ugly duckling to – well, if not a beautiful swan, at least a pugnacious juvenile goose.

Today, the remote descendants of 1980s Windows and MacOS are still leading platforms: colossal juggernauts, with big ad budgets that try to persuade you that they were always meant to be that way. They weren't, but as Helmuth van Moltke said, no plan survives first contact with the enemy.

Windows 1 was a dismal failure, Windows 2 was mostly a total flop but did have some redeeming features, and Windows 3 was perhaps the greatest comeback story in the history of software at that time.

(To be honest, Kalinina's article caught our attention right at the start, with an image from Ranma ½, this vulture's favorite manga. On Kalinina's Mastodon account, she posted some threads of bloopers, including a mention of that particular image, which hints at the amount of work it took to obtain that screenshot. As former Windows 2 users ourselves, we empathized deeply with her angst at Windows 2 memory management. We recognize a labor of love when we see it.)

Some of the serious erudition here is displayed in easily missed details such as image captions, such as this masterfully understated description of Windows 2's text editor:

These were not the good old days. Windows 2 was a nightmare, but at least it didn't have a built-in automated plagiarism generator. We also appreciate the nod to the fact that applications minimized to icons on the desktop, and this was normal for 1980s GUIs. The Reg FOSS desk has lost count of the number of earnest youngsters who have told us this is "proof" that Windows 2 had a taskbar. Presumably the closest these people have ever come to Windows 2 is screenshot galleries, but the screenshots stand as evidence that Ms Kalinina really did get this stuff working, and very recently too.

There's more. For instance:

Well, somewhat, yes. Again, though, it's the history that's interesting. Microsoft combined Calendar and Cardfile to create Schedule+, then combined Schedule+ with Microsoft Mail – called Network Courier before Microsoft acquired it in 1991 – to create Outlook. So, yes, these tiny apps were more useful than they looked ... although more so to Microsoft than to you, especially today.

Some of the other picture captions are also nod to Microsoft's long history of acquisitions, such as that of a now-obscure presentation app:

With some of the other comments, we're not completely sure that they're pointers to relevant history or not. For example:

This is true, but it hides deeper truths, some of which are very relevant to OS/2. So, yes, the networking in Windows 2 was non-existent, but that's partly because it ran on DOS, and the networking in DOS was also non-existent. At one point, that was a killer advantage of OS/2, and Microsoft sold the OS/2-based LAN Manager well into the Windows 3.x era, until Windows NT was ready.

But there's context relevant to the competitive landscape here. Microsoft grasped the importance of peer-to-peer networking relatively early. The first Microsoft client OS with it was Windows for Workgroups 3.1, released in 1992. That's the same year that IBM shipped OS/2 2.0, with no networking at all built in. (If you wanted that, you could add it, certainly. Here's a contemporary guide, which is about 15,000 words long and includes pricing. Yes, it cost extra.)

(Oh, and that terminal program? Back then, it could only talk to a serial port to let you drive a dial-up modem by hand, using Hayes AT commands. The name wouldn't be reused for a terminal emulator for the Windows command prompt for another 34 years yet.)

There's more relevant competitive context in here as well. Kalinina's article lists some of the commercial applications available for Windows 2, saying: "This is more to illustrate that Windows 2 had some useful apps." – with the bonus of more snark at LLM bots:

The overall history context here is in several parts. It shows what Windows looks like at various points in time, and also shows you what rival graphical desktops look like. It is now 40 years since Windows 1.0 came out, meaning that there are middle-aged adults who've worked in the computer industry for their entire lives who've never seen this stuff. A quick look at screenshots of Windows 1 and Windows 2 gives a misleading impression. They're both so ugly that at first glance, they're quite similar. In fact, they aren't. Windows 1 only had tiled windows, but Windows 2 had overlapping ones. Not only does Kalinina illustrate this, she also explains why:

Showing this kind of history, with examples and illustrations, matters. Most people don't remember this period – if they're old enough to have been there, then they're old enough for memories to fade. The key points in the timeline are laid out.

1985

The relevant heading: 1985-11 Microsoft Windows 1.01. That's the same year that Intel launched the 80386 chip. Windows 1 couldn't usefully use the 386's fancy features. It couldn't even use the 80286.

1986

It is not looking good for our hero: 1986-09 Windows is on life support?

1987

At last! 1987-09 Windows/386 (2.01)

And we get a cold and unromantic accounting of the failures of several of these systems. It points out that Windows/386 came out before Windows/286 – nearly a whole year before, in fact.

But Microsoft did catch up. The 386 edition of Windows 2 could in fact multitask DOS applications rather well – and Kalinina points out when and how, in the section titled 1987-09 Windows/386 (2.01).

Hidden under spinners is some good solid tech info too. For example, if you click the triangle next to Windows/386, preemptive multitasking and real mode memory management, you will find a detailed description of the multitasking that was present even this early on. Under Wait, why does it look so ugly? there's a highly instructive discussion of color and font handling.

Scroll down a bit to the section on Windows/286 and there's more hidden under HMA: the 286 part of Windows/286. Here, The Reg FOSS desk has to raise his hand and admit that he still badly misunderstood this stuff just a few years ago. He's not alone. Even the great Raymond Chen got some of this wrong. (For the critical – and critically important – comments, see the archive.org version.)

So, for instance, here we learn that, despite its name, Windows/286 was not, in fact, 80286-specific. You can launch it on an 8086 and it works. This means that Windows/286 didn't have anything akin to "Standard Mode." It couldn't use all 16 MB of RAM the 80286 could address. In fact, all that Windows/286 could use was the 64 KB "high memory area" that DOS's HIMEM.SYS driver made available. The bottom line is confident, to say the least:

This kind of detective work is very hard to find. Almost nowhere still has this kind of info, and without it, it's hard to see through even historical marketing piffle. The result is conspiracy-led nonsense like "Microsoft intentionally sabotaged OS/2 because it wanted to sell Windows!"

1988

This is a big year. Stuff gets real ... and protected, and enhanced.

1988 is the same year that Intel delivered the second member of its 32-bit 80386 family. The cheap, cut-down 80386SX version, with a 16-bit memory bus for cheaper motherboards and cheaper RAM, appeared in June 1988.

Also that month is the beginning of the real history of how Windows 3.0 started to exist, which you find under the heading 1988-06 Windows kernel for the protected mode. What became Windows 3.0 was an internal skunkworks project, led by ex-Microsoftie David Weise.

Some of this story is still online in places – if you know where to look. For example, some ancient Microsoft blog posts, such as "I Found a Cool Little Problem That I Just Couldn't Resist Solving," and "Farewell to one of the great ones." Weise was ably assisted by Murray Sargent, who added crucial support to his debugger. We don't know anywhere else that brings it all together in one place like this, though.

Later that year, and half a dozen sections of this epic after the debut of Windows 2, we get to 1988-11 OS/2 1.1. Considerable time passed before a feature-complete version of OS/2 with the Presentation Manager GUI finally arrived.

And it still couldn't multitask DOS apps ... although a version of Windows that could do that shipped the year before. The chronology of the development of GUI OSes, and the way that Windows gradually and partially caught up, with the Mac, and with GEM, and with apps, and with 32-bit PCs, is laid out. And it's also a timeline of OS/2's failure to keep up with the pace of technological development.

Always remember Hanlon's Razor

Between Windows NT and the equally vast Linux – 1980s monolithic proprietary Unix, reimagined and reimplemented as FOSS – almost everything in the world runs on these huge software stacks. It's almost impossible to imagine them changing course, or anyone dumping them for alternatives. In fact, though, the industry has done this multiple times, including in the last three or four decades ... but the problem is that nobody was writing down what happened. Thanks to Nina Kalinina, you can find some vital parts, all in one place. It's just that it's ... on the long side.

Microsoft really did have good reasons to believe in OS/2 in the 1980s. MS-DOS was a horribly constrained, limited operating system, and early versions of Windows weren't much better. Windows 1 and 2 were deeply flawed, but Windows 2 did have a few advantages, which are laid out here, such as overlapping windows.

The industry desperately needed something better, something that could use the megabytes of RAM that late-1980s PCs had, and for some years, it really did look like OS/2 was that something. But by 1988, it's apparent it wasn't. At the end of the same year that the 386SX shipped, the second release of OS/2 was finally feature-complete. It delivered a GUI, but no DOS multitasking. OS/2 1.x let native apps use the extra memory of a 286, but it delivered no benefit to DOS apps.

No, Windows wasn't some elaborate plot. The history clearly lays out the very serious limitations in all of Windows 2's three separate editions. Windows 3 was planned to be the end of the line, just spruced up with a lick of paint borrowed from Presentation Manager in OS/2 1.2 (released in October 1989).

The final heading of this epic history relegates the real next step to a post-postscript: P.P.S. 1990-05 Windows 3.0.

In 1990, Microsoft released the end result of a secret project by just two engineers who managed to get the most important core functions of OS/2 working in this DOS-based GUI. Thanks to David Weise and Murray Sargent, Windows 3.0 was the first version that could not only access but run apps in up to 16 MB of RAM on an 80286. That equaled OS/2 1's main selling point.

And if you had an 80386 – even one of the cheap, cost-cut 386SX machines – Windows 3.0 delivered hardware-assisted multitasking DOS sessions and true virtual memory – as well as all the 286 goodies. OS/2 1.3, released at the end of 1990, still couldn't do that. (If you want to know more about how Windows 3.0 delivered this, Kalinina links to a deep-dive Inside Windows 3.)

It was not Microsoft's machinations that sank OS/2. Microsoft's management didn't know what Weise and Sargent were up to until they demonstrated it.

As we said last time, it wasn't OS/2 2's ability to run Windows apps that critically weakened it. Version 1.x was already the flop. It wasn't the lack of native apps, OS/2 1 did have apps. For instance, the native OS/2 version of the industry-leading word processor, WordPerfect, shipped in March 1989, and the native OS/2 version of the leading spreadsheet, Lotus 1-2-3/G, shipped in March 1990 – two months before Windows 3.0.

Microsoft didn't sucker those industry partners into doing expensive R&D as a feint. It was doing the same itself. In 1990, Microsoft Word 5.5 was a "family app" that ran natively on OS/2 as well as on DOS. Excel 2.0 (and 2.1), and Word for Windows 1.0, all predated Windows 3.0. Microsoft had to put out updates for both to get them working on the shiny new Windows – but those updated versions, Excel 2.2 and Word for Windows 1.1, both also shipped as OS/2 versions. Even that late, the company was hedging its bets.

The key thing to take away from the history of Windows, and Microsoft's subsequent industry-dominating success, is not that Microsoft planned it. It didn't. There was no long-term corporate strategy here, no tactical investment into Windows engineering to make it more competitive. Microsoft got lucky – and then did one thing IBM was lousy at. It acted fast – just as it did to land the IBM deal for PC DOS a decade earlier. ®