Embrace, Extend, and … Copilot? (Premium)

The phrase “embrace and extend” was long synonymous with Microsoft, but it’s also widely misunderstood and unfairly vilified as a strategy. Worse, it’s been mischaracterized as “embrace, extend, and extinguish” and “embrace, extend, and exterminate,” largely by Microsoft’s competitors. Which makes sense, in a way, since “embrace and extend” isn’t even the version of this phrase that kicked off the controversy in the first place.

“Embrace and extend” allegedly originated with Bill Gates. But it was Gates’s regular use of this term that caused Microsoft engineer J Allard to riff off it for his seminal 1994 internal memo, “Windows: The Next Killer Application on the Internet.”

He called it “embrace, extend, and innovate.” And this memo was part of a plea to get Bill Gates and the company’s other leaders to wake up and address the competitive threat posed at the time by the Internet and its newest and friendliest interface, the World Wide Web.

He recommended a phased approach.

“To build the necessary respect and win the mindshare of the Internet community, I recommend a recipe not unlike the one we’ve used with our TCP/IP efforts: embrace, extend, then innovate,” he wrote. “Although we are years behind many of our competitors on the Internet, our agility and creativity will allow us to catch up quickly.”

In the first “embrace” phase, Microsoft would need to understand the “strange culture” of the Internet, with its roots in UNIX, to “determine the needs and the trends of the user base” so it could deliver products that met the community’s needs. Key among them the coming versions of Windows, 95 and NT 3.5x, and new Internet navigation tools and servers.

In phase two, “extend,” the company would “establish relationships with the appropriate organizations and corporations with goals similar to ours” and deliver “well-integrated tools and services compatible with established and popular standards developed [by] the Internet community.” Here, Microsoft would embrace the (then) current Internet technologies in Windows and integrate standard Internet tools and services into its base products.

And in the final phase of this “Internet onion,” as he called it, Microsoft would need to “innovate” to move into a leadership role in the creation of new Internet standards. It would create “off-the-shelf” software built with “Internet awareness,” which probably sounded pretty radical at the time. And Windows would become “the next-generation Internet tool of the future,” fulfilling the title of Allard’s memo, “the next killer application for the Internet.”

None of that sounds even remotely malicious to me. Indeed, the “extend” phase sounds like more another round of “embrace.” In effect, he was talking about Microsoft becoming part of a community that already existed and then eventually assuming a leadership role that respected that community’s customs and history. Yes, phase three represents a shift from UNIX on the backend to Windows NT, and, yes, Windows becomes the default client on the front-end, but not through nefarious means. Just competing by being better.

The history of what happened next is well-understood: The J Allard memo wasn’t enough to get Bill Gates to take the Internet threat seriously, but Gates finally did wake up later in 1994 and, well. You know. Bad things happened. If you’re unclear on the history, I cover this period in Windows Everywhere and in the Programming Windows: The Internet Tidal Wave (Premium) article that later became part of that book.

Great. But why discuss this now?

“Embrace and extend” is one of several topics I wanted to revisit in the wake of reading (and re-reading) and reviewing Steven Sinofsky’s book, Hardcore Software: Inside the Rise and Fall of the PC Revolution. When I first dove into that title, I found some insights that surprised me, and I originally planned to detail several of them in my review. And then it occurred that these might make sense as separate topics. This article is the first of what I hope will be several similar deeper dives into topics that I think require more space than a quick mention in a book review.

I started with this topic because when I thought about “embrace and extend”—really, “embrace, extend, and innovate”—again with the benefit of additional hindsight, I discovered something fascinating. Microsoft is arguably using a modified version of this exact strategy today with Copilot and AI. And it could not have telegraphed this strategy more clearly.

I have repeatedly referenced Stevie Batiche’s stunning May 2023 appearance at Build 2023, during which he explained how Microsoft would implement AI across its stack, most recently in Windows 11 is About to Get Its AI Moment (Premium). It would do so using three application structures that would help Microsoft and the industry transition from the traditional software world of the present to the AI future. A three-phase strategy, if you will.

The first structure is called AI beside because Microsoft would implement AI experiences called copilots that would visually sit beside existing (legacy) apps. And we’ve since seen this exact implementation in products like Copilot in Windows 11 and Copilot in Microsoft 365, in apps like Word and Excel.

The second structure is called AI inside, where Microsoft would deliver new experiences that look and work like evolved traditional software but are architected around specific AI capabilities. Batiche cited Clipchamp and Designer as examples of this structure.

And then the third structure, called AI outside, requires a total infrastructure transformation and is where this all comes together. Instead of running individual apps, we will use AI to orchestrate multiple applications and services on our behalf to complete discrete jobs. A copilot of copilots if you will.

AI beside, AI inside, and AI outside. Embrace, extend, and innovate. Two different plans, from two different eras, with the same goal: To move the Microsoft stack of the day from its traditional, familiar past to an exciting but unfamiliar future. This could be called, “Add, integrate, and rebirth.”

Don’t overthink this: No comparisons like this are ever perfect. It’s not like comparing two DNA transparencies and seeing an exact match. And in this case, I will point out that Allard’s 1994 plan was literally three steps that needed to happen in an exact order, while Batiche’s explanation describes three implementations that overlap. For example, we have “AI beside” and “AI inside” apps running side-by-side today and probably will for a long time. We didn’t move from one to the other.

And to be clear, the similarities are not purposeful, not an indication of some secret long-term strategy that demonstrates a greatness we didn’t realize was there all along. It’s just a coincidence. Or, more likely, two examples of an inherent similarity, conditions that exist, on some level, during any platform transitions like these two. Whatever, it’s still fun.

“Embrace and extend” exploded into the public consciousness because of Microsoft’s antitrust issues in the late 1990s, and various versions of this term popped up repeatedly in testimonies and public court records. Intel’s Scott McGeady is likely the source of “embrace, extend, and extinguish”: This hostile witness testified that Microsoft executive Paul Maritz used this term in describing how the company would defeat Netscape and usurp Internet standards. And two other Intel executives, Russell Barck and Rob Sullivan, credited the similar term “embrace and smother” to Mr. Maritz too.

Whether he said any of that doesn’t really matter: Aside them being hearsay, we were perhaps a little too sensitive to the tough, competitive language we discovered Microsoft’s executives using in the late 1990s, a time at which the company still operated like a scrappy street fighter and didn’t quite understand its own market power. The real problem was what Microsoft did by artificially commingling Windows and its web browser to defeat a competitor. Or, as Maritz allegedly put it, by creating “various levels of dependencies between the operating system and the browser that would differentially advantage their browser.” It would “cut off Netscape’s air supply.” There’s that language again!

But in Hardcore Software, Sinofsky provides another, less violent view of “embrace and extend,” noting that it was never meant as a loaded expression, and was instead just the way things were done in the industry at that time.

“Computing evolved by companies constantly finding new ways to extend existing software and hardware, elements viewed as commodities, in unique ways, while at the same time always bowing to the existing investments of customers by claiming an embrace of any existing standards or winners,” he writes. “That was how value was created—everyone built on what came before. That was how everything worked. In a sense, every new product somehow related to the product it would supersede while extending it in uniquely valuable ways.”

“As it would turn out, embrace and extend is precisely how every open source innovation was commercialized, particularly as software moved to the data center. Not everyone believes that is good, but that is what happened.”

Indeed, how can you look at Netscape Navigator, the product that kicked off this controversy, and not see how its makers embraced what had come first with Mosaic and extended that, in part by commercializing it at scale? Choosing a more current example, this is a key tenet of the Apple playbook too, where this company seems to enter markets late but succeeds by adding value and nailing the user experience. Embrace and extend, baby.

And besides, J Allard’s “embrace, extend, and innovate” memo had it all backwards anyway, Sinofsky claims.

“The Internet was larger than Windows and could not be contained by an operating system,” he writes. “Windows was an application on the Internet.”

A killer app, if you will. Another term so many misinterpret. And another key difference, perhaps, between “embrace, extend, and innovate” and “AI beside, AI inside, and AI outside.” It’s not clear that any one AI capability will emerge as a singular killer app, or key experience. Instead, it appears that this wave will be defined by hundreds or even thousands of individual advances, each vertical in nature. Perhaps they can be orchestrated by a copilot of copilots.

We’ll see. That history has yet to be written.