Windows 11 collects data on which coding agents and dev tools you use
44 points by viccuad
44 points by viccuad
I really wish we had user privacy laws in the US that actually protected us from egregious privacy invasion like this. Reading this reinforces my decision to switch from Windows 10 to Debian last year.
Microsoft Windows is an excellent spyware just lacking a decent operating system.
It is important to understand that Windows exists solely to spy on you at this point. Microsoft has joined Google on the wrong side of "Don't be evil."
This sounds like an unawareness of the history of Microsoft if you think they "joined" anyone in that, rather than being the people who pioneered it.
I was trying to be gracious.
I know Microsoft pretty well, and used to live on 156th in Redmond only a quite-literal stone's throw from billg's office (and before that on 148th in Bellevue only a stone's throw from billg's old 520 campus office).
Naive take that doesn't survive even a cursory analysis. The world is not a black-and-white place, Microsoft is not a monolithic entity.
If I was looking at a crash report or similar, I’d certainly appreciate knowing if the machine in question had a coding agent running wild, because it’s very likely the machine is in a weird state because the agent is doing something bizarre to work around some problem it’s encountered.
Maybe it’s still a bug, but quite different from a bug that happens to a normal user just going about their business.
There’s certainly a privacy trade-off here, but there are definitely non-malicious reasons to be collecting this.
This reminds me of the dynamic with dual-use technology. (The irony being, of course, that AI itself is also dual-use!)
I can understand that some companies and organizations within the US are using Windows, but for privacy and security reasons I can't fathom why anyone else would.
Is it because of salespeople and lobbying, or pure technical inertia?
It's a good operating system that is both deeply customizable and tends to "just work" in its default state.
That depends on the use cases. Windows is worse than Linux for many use cases. Which use cases do you have where giving up on privacy and sovereignty is acceptable? Gaming?
Which use cases do you have where giving up on privacy and sovereignty is acceptable?
Printing. Good grief, printing. I am losing my mind trying to print on Linux with certain printers. Some are well supported, others... a struggle to put it mildly. Before I personally buy devices like printers, I briefly research to make sure it works on Linux.
Fair point! Over time, I hope we can get rid of all printers, except for artsy plotters. Until then, a dedicated Windows (or macOS?) print server does make sense, I guess.
The biggest thing remains compatibility with certain software that only runs on Windows. Most people don't care as long as they can effortlessly run all of their software. That especially includes businesses that have expensive licenses for software that's essential to their operation.
That's one of the reasons we need FOSS that both fills the gaps and expands the software available to users of free operating systems to make them appealing.
One issue is that the network effect tends to go one way. FOSS tends to care more about compatibility, and even if the main developers don't, anyone else can port it. That's a major benefit of free software, but does often put us in a catch-up game instead of being able to get ahead.
Nicer system administration and scripting (great settings UI and PowerShell is a much nicer shell than bash/fish/zsh with good system integration)
This is all subjective. I find the Windows Settings UI to be wildly confusing, cluttered with things I don't need, and organized poorly. I also find PowerShell to be overly verbose.
Windows API is quite nice (and it's the one mainstream API preventing a desktop POSIX monoculture)
Again, taste. but I will give points for Windows keeping a POSIX monoculture from taking hold.
.NET is a great development platform that's still Windows-first
WSL2 makes cross-platform development much nicer than from Linux
Both subjective. It's things like the data collection from TFA, the advertisements, the constant attempts to upsell me on Microsoft services (including pointing directories in $HOME to OneDrive, which is just wild) that keep me from thinking Windows is at all "nicer" than any Linux desktop.
Windows are a single platform with stable APIs, making software distribution significantly easier (and making it easy to get new software)
APIs are nice and stable, fair. But "making software distribution significantly easier"? "Easier to get new software"? The Windows Store is a slow, cumbersome, nasty experience, especially compared to a decent package manager on Linux.
Desktop security is taken much more seriously
Than what? I mean, granted, Microsoft is huge and they do care about security incidents. And, working for Microsoft, you'd know better than I do what their internal state is. From the outside, it's harder to see.
Windows Settings UI to be wildly confusing
This is a strawman. The parent commenter explicitly said 'system administration and scripting', and the context it should be taken in is server and fleet admin, where tools like Active Directory (or Azure Entra ID or Intune or whatever they call it now), gpedit, and, of course, PowerShell absolutely shine. And for the record, KDE's, GNOME's, and now macOS' settings broadly use the same 'topics on small left pane, details in main pane' UI.
I also find PowerShell to be overly verbose.
I am very tired of this argument. People take one look at Get-ChildItem instead of ls and say 'too verbose!' (read in SpongeBob script for effect). And then they dismiss what is a damned good scripting and infra language.
Firstly, there is a logic to every cmdlet: it is Verb-Noun. For interactive use, there are built-in aliases for frequently-used cmdlets, and interactive use accepts mixed-case. Most parameters are -Named, and there is a logic to this naming as well. These parameter names can be shortened to the first n letters, again, for interactive use. So one can absolutely write
ri -r -fo $env:localappdata/vcpkg
instead of
Remove-Item -Recurse -Force $Env:LOCALAPPDATA\vcpkg
PowerShell's cmdlets all emit rich objects which can be introspected, parsed, and pipelined. It completely obviates cut, awk, sed, grep, cut, tail, etc. It has native JSON handling, and therefore also obviates jq.
Scripts, cmdlets, and functions have type-checking for declaring parameters. This then auto-generates help documentation and autocomplete. Functions have sane error-handling and can even throw exceptions, instead of returning some arbitrary unsigned short value.
All of these mean that PowerShell scripts can be maintained more easily. POSIX scripts, by their extreme brevity and honestly startling syntax are 'write-only. Anything above about 100 lines starts to become very difficult to handle, with set -ex and miscellaneous hacks required.
Since it runs on the CLR, it can natively call out to .NET methods. This also means that we can write a library with heavy business logic in C# or F#, produce a DLL, and write a wrapper script over said DLL in PowerShell and directly call out to methods in that DLL; no main function with args required. And of course, one can put a PowerShell script inside a .NET debugger and step line by line, column by column, just like C# and F#.
PowerShell even has a style guide. This all makes PowerShell a scripting platform easily rivalling Python and completely blows most POSIX shells out of the water, whilst also being nice to use interactively. The one negative point with PowerShell is its startup time because it is tied to a CLR; the shortest I could get is a few hundred milliseconds, and UNIX shells, being native binaries, start up in a fraction of that.
But "making software distribution significantly easier"?
For the end-user, binary-first distribution is much easier than source-first distribution. The stable ABI of most Windows libraries means that one can compile for Windows 2000, and generally expect the same binary to run with little fuss on Windows 11. That is 26 years of binary compatibility that is not seen on any other platform. Windows Store is generally an aberration, but again, the infrastructure that underpins it—MSIX—is pretty slick and is competitive with the zoo of Linux package ecosystems.
If one really wants CLI package management, then winget import should work. One can set up a Windows dev environment basically as easily as pacman -S base-devel llvm with winget. There are some rough edges like packages spewing the desktop with shortcut icons, but I think that can be worked around and is in progress.
The stable ABI of most Windows libraries means that one can compile for Windows 2000, and generally expect the same binary to run with little fuss on Windows 11.
You can go even further back if you want - stuff compiled for x86 Windows NT 3.10 (released in 1993) should still run on Windows 11 too. That's 33 years of binary compatibility!
Somebody found an executable from a pre-release version of NT 3.1 that still runs. https://blog.enlyth.com/posts/oldest-windows-gui-app-win-11/ . Also FreeCell, as a Win32s program, still works.
My main problem with PowerShell (as a Unix greybeard who has been required by circumstance to use Windows for a living, mostly writing .NET) is not so much that it's verbose, but that Verb-Noun is the /wrong/ order for cmdlets to be named in. The nouns would function as a namespace prefix if they were first, and if I were looking for something on a VPN connection, I could put VPN first and tab-complete. But no, I have to guess which verb I want (is it Add- or New-?), and then see if there is one that applies to the thing I want to operate on. /Much/ worse than say, nmcli or firewall-cmd, where fish or zsh completion will quickly lead you to what you're looking for.
As for system administration, I /will/ say that the server admin GUIs on Windows Server are much better than the desktop settings GUIs on Windows desktop. And Windows /could/ be deployed as a much better operating system than it conventionally is. For example, NTFS is a really good filesystem... until you bog it down with filters that make normal things like git status unbearably slow.
Verb-Noun is the /wrong/ order for cmdlets to be named in
Yes, I dislike it as well. As a workaround, you can type *-Noun and tab-complete to see all verbs, but it would work so much better the other way around (or even go the subcommand route, command get instead of Get-Command).
NTFS is a really good filesystem... until you bog it down with filters that make normal things like git status unbearably slow.
Yes. The first trick is to use a separate data partition because the system drive has a bunch more filter drivers attached. If on win11, make it a Dev Drive (I'm not sure how much better ReFS actually is, but the async Defender mode helps a lot). The second trick (if allowed to on a work machine) is to completely disable Defender using GPO. Suddenly, all of the weird I/O pauses and random slowdowns are gone, binaries run immediately after compilation,...
Actually, I did mean the desktop Settings UI. From what I heard from my sysadmin friends, AD, Entra,... are very good indeed, but I don't have personal experience so I didn't feel qualified to mention them.
This is all subjective
Yes, as are most reasons why you'd pick one OS over the other. Turns out that with 30+ years of convergent evolution, all the big 3 desktop OSes are mostly fine™ for most kinds of work, there are some specific areas where one is better and in most areas, personal preferences play a big(ger) role.
It's not a race, I'm not claiming that Windows are the best OS in the world, I use Linux on quite a few headless machines and like it (and support wider adoption of free software in general), I'm just explaining why I personally prefer Windows for desktop work, even though I disagree with some of the technical and ethical decisions.
I find the Windows Settings UI to be wildly confusing, cluttered with things I don't need, and organized poorly.
Entirely possible (although much improved with win11), but with 20 years of muscle memory and intuition, I can find and configure anything I need in a few seconds. Every time I try a GUI-centric DE like Gnome, the graphical settings only provide absolute basics and you have to use CLI or manually edit config files in various obscure formats for anything more specific.
I also find PowerShell to be overly verbose.
There are aliases and IDE-level autocomplete that make this a non-issue for me, coupled with having a much more flexible language that's actually composable, unlike POSIX shells. For example:
ls -Recurse -File | ? Length -ge 1MB | sort LastWriteTime | select Name, Length
(autocomplete works for all of the field and parameter names)
It's things like the data collection from TFA, the advertisements, the constant attempts to upsell me on Microsoft services (including pointing directories in $HOME to OneDrive, which is just wild).
Understandable, I find it inexusable that those are the defaults in a clean Home installation, but with Pro/Enterprise most of that is not really a thing and for the rest, you disable it once and never touch it again. Not great, but also not really much of an issue in practice for a power user.
Generally, I don't particularly care what Microsoft thinks about how I'm supposed to use Windows. It's my computer, I use it how I find efficient, whether it aligns with Windows defaults or not.
making software distribution significantly easier
As a developer targeting Windows, I can compile binaries for 2-3 architectures (which is trivial with MSVC), run them through WiX or a similar tool and ship a release that all of my users can immediately use.
On Linux, I can either publish a tarball and hope that maintainers will build a package for me (lol), or learn how to build packages for 10+ different distros and deal with different dependency versions etc., or publish a Flatpak (or was it Snap? Or AppImage?) and make ~everyone angry. Then, you can deal with Debian shipping a version of your software that's 5 years out of date.
Package management on Linux is so good that most server-side projects are now shipped as a Docker image instead. (only half joking)
Than what?
Desktop Linux. To give a practical example, if you leave a locked laptop with a default win11 installation in a room full of hackers, I'm fairly confident that no data leaves that laptop. If you did that with a default installation of most Linux distros, I'd have it unlocked in a few minutes.
TPM-based disk encryption should be table stakes, same with Secure Boot, lockscreen that actually locks (less of a problem now that Wayland finally displaced the horribleness that was X11), driver signing, elevation process that's non-trivial to bypass,...
Few more I remembered:
SMB shares work pretty well on Linux if you use the kernel driver. gvfs-smb (and gvfs-mtp too) on the other are a mess.
Desktop security is taken much more seriously
I think your point is more that it is secure in the sense that no third-party actor will be able to steal your bank account while on windows. But it's somewhat ironic to say this in a thread about how Windows 11 has malicious behavior built in.
Among the already mentioned things, administration and policy/compliance is far easier on Windows than Linux. Group policy/domains/AD DS doesn't have a great alternative on Linux, and the closest offering leave much to be desired (Red Hat has one iirc but it's definitely something).
It's worth noting that many of the privacy issues are defaulted off, easy to disable, and/or easy to redirect to your local admin instead if you're on windows enterprise. Companies pay to not be spied on, consumers don't get the choice.
I think it's highly, highly likely that companies are spied on, regardless of payment, especially if they are outside of the US. Why should they not be? Microsoft and the US has both the means and the incentive. And apart from completely stopping all network traffic from Microsoft products (including Teams), it's hard to tell both what's going on in the code and also what's being sent over the network.