I thought multiple times over the last month about Windows Recall vs. WinFS. But not in the way that you might think. As Windows Recall went from overhyped feature to almost gone, I couldn’t stop comparing the two.

But again, not in the way that you might think. I pondered creating a list of similar traits to these two overwrought pieces of infrastructure sold as life-changing.

However, WinFS and Recall aren’t the same thing, and they weren’t intended to do or be the same thing. But they have some similar problems.

You can compare a few fundamental traits of the two:

1. The benefits of both were overhyped and misunderstood by press and pundits
2. The real costs of both and their intrinsic problems were never made clear
3. Both Recall and WinFS were Spruce Goose projects
4. It seems apparent that executive pressure drove them when rational thought would have shut it down.

Let me explain each of these in a little bit more detail.

The benefits of both were overhyped and misunderstood by press and pundits

Like so many features of Longhorn, WinFS grew to appear immensely more important and more pivotal than it actually was. People fell in love with the idea of an object-oriented filesystem, whatever the hell that would mean in practice, and every time Microsoft insisted on trying to get this rocket airborne, it blew up. In Microsoft classic form, hyperbole was stated so many times that it became pseudo-fact.

But in the end, WinFS played a huge role in Longhorn dying on the cutting room floor in 2004 when the entire OS was reset off of the Windows Server 2003 Service Pack 1 codebase. The Windows Server team had made real security changes, shipped a great service pack, while shipping support for the AMD64 processor that would keep Windows alive into the future, while client-focused Longhorn had plowed into a wall.

However, the most important thing you need to remember about WinFS is that:

WinFS
Is
Not a
File
System

That’s right – it never really was a file system. The WinFS project really lived under the SQL Server team, not a team in Windows. I sat through countless meetings with WinFS team members, and tried to listen to their scenarios. In particular, WinFS was an instance of SQL Server duct-taped to the OS, with a Windows Explorer that had been maimed of all legacy navigation paradigms (because you wouldn’t need them anymore, and everything would just look like a fileshare). It was literally the worst of all worlds for people who needed to use Windows day to day. The OS was a dog because you’d shoved a SQL Server deep into it, and the shell was unnavigable in the interest of making everything “easier”.

The costs to us in the Windows setup team for WinFS in Windows Longhorn were huge. It slowed down install, made first boot slower, and reduced performance once the desktop finally arrived. And for what? Everything as an object? What the hell does that even mean in the end?

In a meeting with a WinFS PM, frustrated with the fact that the real file system team didn’t want to work with WinFS, I referred to their project as “a strap-on filesystem”, which it was. WinFS was not a file system, it was a malignant tumor on the back of NTFS. Sorry, not sorry.

In terms of Recall, I feel the same way. It seems almost embarrassing that a company that not only makes their own Chrome-based web browser, but makes the entire OS underneath it, could have come up with a way to track user activities without this horrific brute force of “take a bunch of screenshots and then use OCR to see what the user was doing.

It’s gross. It’s offensively thick. It’s like using a Sawzall to cut a tomato. But I think Microsoft went this way to both demonstrate the concepts of their ML infrastructure on PCs, while likely also coming up with some horrifically gory “telemetry” or ML training data that should absolutely not be collected.

Again, Microsoft owns the OS and owns the browser. I’ve heard some people say, “I want to be able to find the thing I was searching for earlier today.” All I can tell you is that you can do that across Apple devices without needing to nave screenshots and OCR underneath it. Why does Microsoft need to go this route? More importantly, that avenue can never work – or can never work safely – across multiple Windows PCs. So if you can’t remember if you searched for the thing at work or at the office, you’re going to have to search both. That’s a pretty huge drawnback for the anvil-based approach that Recall takes. Why on earth did Microsoft have to take such a brute force approach with Recall?

The real costs of both and their intrinsic problems were never made clear

The problems of “embedding” an instance of SQL Server into the OS, and then building in this hyper-aggressive indexer (I still remember the performance problems—and cursing—at first boot the week it came into the main builds of Windows), weird, abstract directions for the Windows shell (like dropping drive letters) and then underneath the convoluted story of how it would move your files to a little hoarding cache under System Volume Information and then put a hard link back out to your original file… It was pretty weird. Unfortunately when WinFS was discussed outside of the company, it was discussed as if it was this well-designed thing of beauty, when it was this weird amalgam of the Windows shell, SQL Server Server, and just a sprinkle of actual file system to hold it all together.

But people outside of the company got so excited, because they’d fallen in love with the idea of what WinFS was – which was nowhere near the reality of where it was headed.

I can repeat that same sentence with Recall and have it work too:

But people outside of the company got so excited, because they’d fallen in love with the idea of what Recall was – which was nowhere near the reality of where it was headed.

Both Recall and WinFS were Spruce Goose projects

Both projects felt like they were research projects that got loose. No, not Microsoft Research. Like something an intern would design, without regard to complexity, cost, usability, or the timeframe for shipping something practical, secure, privacy-respecting, and maintainable.

With WinFS, that resulted in a huge contribution to the instability of Windows Longhorn, as well as exponential expansion in the OS internals for no clear benefit to the end user. It was, at the core, a SQL Server-based feature that targeted developers. In the end, Longhorn features had to run a gauntlet and jump over appropriate hurdles to stay in the new builds of Windows after the reset, but long before it became Windows Vista. In the end, it was separated from the OS and would ship later. I knew as soon as it was deferred it was dead; and eventually it was.

With Windows 11, Recall creates different but somewhat similar problems. Microsoft’s “Copilot+ PCs”, which are confusing both in terms of bizarre branding and technically; the fact that no Intel or AMD PC (what most people buy), let alone any Arm-based Windows PCs that people already bought this year, can comply with it or run Recall. Only brand-new, Arm-based PCs with Qualcomm’s latest silicon.

So again, it’s touted as a feature for consumers, but come on. Recall was demonstrated at Build; it was clearly intended as a feature for developers to get excited about.

So why the hell did Microsoft develop this feature in complete silence without any conversation, and then just plan to shove it out without any testing in public, while denying the fundamental security and privacy problems inherent in this design, until proven wrong in public multiple times? It feels evil, but I’m inclined to invoke Hanlon’s Razor. If any executive at Microsoft knew that Recall was as porous – and poorly designed – as it was, then they have a lot to answer for.

It seems apparent that executive pressure drove them when rational thought would have shut it down.

With WinFS, it was clear that BillG was in charge, and he wanted WinFS to be in the OS. Even when they seemed to be lost in the wilderness, there was significant pressure to do whatever it took to make WinFS happen.

With Recall, it’s hard to imagine that Satya wasn’t a strong internal promoter of it. It resonates solidly with the company’s intense focus on all things “AI” (ML, really), including the Baskin-Robbins collection of things chaotically branded as Copilot (and now Copilot+ too.)

I still don’t get how Recall could make it through countless reviews to almost get out the door as it was, when it was such a… ridiculous? concept.

If Microsoft does ship Recall, and it’s still using ML internally, I’ll be disappointed first, and surprised second.

What’s to come?

We will see if my list above can be amended at some point with one last item:

1. The benefits of both were overhyped and misunderstood by press and pundits
2. The real costs of both and their intrinsic problems were never made clear
3. Both Recall and WinFS were Spruce Goose projects
4. Executive pressure pushed them when rational thought would have shut them down
5. Both Recall and WinFS were discontinued before ever becoming generally available*.

*To be continued…