Microsoft says Windows is becoming an agentic OS

[madires]

Just a another reminder (happened in July):

Not sovereign: Microsoft cannot guarantee the security of EU data - https://www.heise.de/en/news/Not-sovereign-Microsoft-cannot-guarantee-the-security-of-EU-data-10494789.html

2. The far more likely.  AI bubble bursts.  The circle jerk financing explodes and the entire stock market tanks with defaulted debts and bailouts similar or worse than 2007/8.

I think so too. And trying to educate people about LLMs and the likely outcome of the AI bubble is a waste of time as only a very few listen. Instead nearly everyone is throwing more money at 'AI is the future' and 'we can't fall behind' projects and data centers. Investing all that money in sustainable energy would be much more reasonable and helpful.

Clarification: LLMs have their use cases, but they are neither an AI nor a universal solution for all problems.

[paulca]

The jurisdictional data issue is basically because AI farms have very sloppy data governance in how, where and why they collect data.  Let's say the wide and deep net they cast is not very specific about "regions" and GDPR or any DPA.

They claim that what is publicly available is valid training data.  Most of copyright laws and state data protection acts stand against them.  This is only one small chip in the armour that a few strategic law changes could cripple LLMs.

LLMs and other AI neural network processing engines can and do do great and impressive things.  However, almost 1:1 they do some absolutely gormlessly stupid things.

For every 1 AI success story there are 100 failures.  When asked directly, after a "AI is everything" speech an exec team member was asked a simple question.  "I see lots of innovation "lab" work and a lot of talk and PoCs, but has anyone actually bought any AI solutions and have we published any white papers on those successes?"

"No."

[Karel]

Separate agent workspaces and agent user accounts, with security as "top priority"

Redmond stresses that Windows 11's agentic AI features follow a "robust" set of security principles.

https://www.techspot.com/news/110306-microsoft-explains-how-windows-11-become-agentic-os.html

[exe]

I also use Linux to create executables for those platforms (Suse, Mint, Ubuntu, CentOS) that I support but this is under a VMware virtual machine running on a hefty Windows server. And every time I need to update a linux version, I have to spend a couple of days compiling Qt, figuring out what shared libs I need to download, making sure I can share files (this is of course a vmware issue but I have no problem sharing files between Mac and Windows). Oh and OpenGL support for anything other than Windows can be tricky.

Sounds like CI/CD based on docker would solve most of your problems. Including automation. I do CI/CD for one of our projects (https://fireflyzero.com/), and Linux is the easiest one to build binaries for as it runs in docker. With windows and MacOS it was much harder to get everything working, though I'm not a Mac admin, may be I miss something.

[paulca]

I also use Linux to create executables for those platforms (Suse, Mint, Ubuntu, CentOS) that I support but this is under a VMware virtual machine running on a hefty Windows server. And every time I need to update a linux version, I have to spend a couple of days compiling Qt, figuring out what shared libs I need to download, making sure I can share files (this is of course a vmware issue but I have no problem sharing files between Mac and Windows). Oh and OpenGL support for anything other than Windows can be tricky.

Sounds like CI/CD based on docker would solve most of your problems. Including automation. I do CI/CD for one of our projects (https://fireflyzero.com/), and Linux is the easiest one to build binaries for as it runs in docker. With windows and MacOS it was much harder to get everything working, though I'm not a Mac admin, may be I miss something.

This.

Don't even need to go to the lengths of a CI/CD pipeline.  Just docker for a start.

Docker containers are mini-os's.  One for each component in your system.  Instead of configuring the host OS to support your application, all the host OS needs to do now is support Docker containers.  The QT libs are in your docker container along with their deps.  They can be any version you like.  They are somewhat coupled by architecture and running "Windows" in docker isn't really viable.  Running docker in windows is though.  We run enterprise docker containers on Windows 11 all the time.  The target is Kubenetes on linux though.  Depends on language etc.  QT Might be a challenge.

Of course...  docker containers with UIs running on desktops is possible, but not exactly the normal.

[chilternview]

Sounds like CI/CD based on docker would solve most of your problems. Including automation. I do CI/CD for one of our projects (https://fireflyzero.com/), and Linux is the easiest one to build binaries for as it runs in docker. With windows and MacOS it was much harder to get everything working, though I'm not a Mac admin, may be I miss something.

Yes I've heard of Docker but never really looked into it, thanks for reminding me. I should look, as VMware was always a bit flaky and it's unlikely to get better now it's under Broadcom's thumb. Presumably you need to build it from the disto and all the shared libs you need? That last point is where I have the biggest problem with building a linux version of Qt: generating the Qt xcb libs needs all the right shared libs, else it silently fails. And Qt doesn't tell you what libs you *do* need, and it can change every version or so.

In comparison Windows needs openssl, llvm, python, perl and ninja but that's about it.

[paulca]

Here is the docker build for my python microservices.

It starts with a base image called "python"  version 3.  This gives me a micro-linux with Python3 installed on it and nothing else.  Pulled off the internet.

Then I set the current direction and copy the source code into the /code folder on the image.

I run an "Apt update"
I install "iputils-ping" as I use "ping" from python.
Then I take the contents of the projects python dependencies declared in "pip.txt" and install those.
I set the python path to "/code" (.)

Code: [Select]

FROM python:3
WORKDIR /code
COPY . .
RUN apt update -y && apt install -y iputils-ping
RUN pip install -r pip.txt
ENV PYTHONPATH .
~               

To run it you can do something like:

docker theimage:latest --entry-point="python3 /code/thepython_script.py"

and the docker image will run as it's own OS entirely until that script exits when the whole container will exit.

While this uses mostly "python deps" via "pip" the same thing works for a list of "apt install" instructions.

Also note that this image is used for a dozen different "applications", they only differ by the code that was in the direction you ran it and what you set the "entry point to".

Docker is great for running 1 container.  When you have 2 or more that depend on each other there is docker-compose which allows you to manage groups of containers in a single yaml.  Up from there and you get into the full CI/CD pipelines, container registries and eventually Kubernetes.

EDIT: When learning docker, do look up how to get a shell in a running docker.  This will help you "see" from the inside why it's not working.  When doing this, you can, of course, install your favourite tools.  Like "vim" or "nano".  Just add those install commands to the docker and it will contain them next build. 
Additionally when developing the code you can use "dev images" which import the code from the host, so your docker keeps running, but your application changes as you edit and recompile/build the code on your dev machine outside of docker.  This avoids the need to rebuild the docker image in dev constantly.

[madires]

And the harsh reality:

Microsoft yanks Copilot ad after AI fails at a basic Windows 11 task grandma could spot - https://www.windowslatest.com/2025/11/19/microsoft-yanks-copilot-ad-after-ai-fails-at-a-basic-windows-11-task-grandma-could-spot/

[exe]

Presumably you need to build it from the disto and all the shared libs you need?

@paulca shared a nice example. So, I won't try to make "yet another docker tutorial", rather explain what it does and best practices.

So, Dockerfile is a set of rules to build an image. If done correctly (e.g., all packages are pinned), this gives 100% reproduceable and automated building of environment. This image, once built, is cached, so re-building if there is a change, or using it is faster.

The base image can be virtually any linux distro you need. If you need debian, you put "FROM debian:bookworm" (or any other version you want from https://hub.docker.com/_/debian). Same thing for any other distro. I recommend pin the distro version, at least major version.

[chilternview]

Thanks guys. I will take a look.

It seems that Docker is more like an app running in the host OS than a virtual machine, is that correct? Can one run the code you have compiled complete with tests (mine uses OpenGL drawing)? I'm still not quite clear exactly what it allows you to do.

[exe]

Thanks guys. I will take a look.

It seems that Docker is more like an app running in the host OS than a virtual machine, is that correct? Can one run the code you have compiled complete with tests (mine uses OpenGL drawing)? I'm still not quite clear exactly what it allows you to do.

Ah, yes, you can run the code. But with graphics it's a little bit tricky. Do you need headed or headless mode? Headless is when you don't have real screen to output, but rather a buffer. At the end of the test you can get screenshots of your application for further analysis or troubleshooting.

Headed mode is also possible, though I don't know how well it works if you host OS doesn't match container OS. I mean Win+Lin or MacOS+Lin. I'm pretty sure there is a solution, but it may be a can of worms. Even on Linux it's a little bit tricky. I suggest reach out for help on a Linux forum or stackoverflow or something.

The overall pattern is: you build image with all the dependencies, etc. Then run the image (producing container) where you run building of your application. Again, better ask for help on a relevant forum, as there are several options available.

[m k]

not sure why anyone would still agree to use this steaming heap of crap nowadays. apparently lots of folks simply don't possess enough brains to care.

No need to insult anybody. Most people use windows because they are forced to use it. Not because they want to use it.

Again, not true.

I have a different impression.
I'd say that most people don't really care what the actual OS is, they just need that something on it is functional.

Can't remember paying M$ after W3.11 floppies, but would be quite a hassle staying out of the ecosystem.
(customers are in)

It's also possible that being in EU means something, different rules.

[shapirus]

It seems that Docker is more like an app running in the host OS than a virtual machine, is that correct?

Docker containers are rather something in between: they can run processes inside them that the host OS will still see and manage as processes (i.e., you can see them via top), but they will be isolated inside their own environment as far as the file systems and network go and won't be able to reach the host OS besides certain known things (such as the /proc file system, for example). It is also possible to set limits on resources such as CPU time and RAM.

Can one run the code you have compiled complete with tests (mine uses OpenGL drawing)? I'm still not quite clear exactly what it allows you to do.

It can run any processes that the host CPU can run. Speaking of OpenGL rendering specifically (assuming output to the monitor), yes, it should be able to, but I think it would need to connect to the host X server with the process inside the docker container acting as an X client, which can be done over IP or over a unix socket, either of them would need to be explicitly made available to the respective Docker container. I'm not sure if it's achievable with that Wayland nonsense, as it never had support for the client-server mode of operation, if I understand correctly. Maybe it can do it via some compatibility mode. Otherwise, a proper X server (Xorg/Xlibre) would be required.

[paulca]

cgroups and others are the kernel virtualisation subsystems behind it.

If you are already familiar with VMs and with chroot, docker containers are somewhere in the middle.  Rather than just the FS being sandboxed a lot of kernel structures are sandboxed too.  From the containers point of view anyway.  You can see in, but it will struggle to see out.  It's not impossible and there are exploits to gain some access beyond the container boundary, but these are usually easily addressed if of concern by forcing all containers to run as "nobody" and use other infra to manage it's actual access.  If you don't care, you can just run them as root.

There are alternatives to Docker.  The container spec is published independant of docker-io.  Obviously it's really what the kernel supports and it's instrumentation.  A popular alternative is "Linux Containers" or LXCs.  They tend to be used slightly differently to docker, but are much the same.  They tend to be more "full on distro" installs.  Like for instance, you can spin up a container and install Ubuntu onto it directly.  In that they LXCs expect a full boot root setup.  Dockers tend to be more a "single process wrapper".  LXCs are more heavy weight and actually "mount" or "bind" a rootfs, not just an ephemeral cache one like docker.   Note a full distroVM will need 2Gb of RAM or more allocated to it 100% of the time.  An LXC running the same distro will probably only use a few hundred meg.

Typically....

Use a docker style container when you want to run an isolated process.
Use an LXC style container when you want to run a monolith "trad" server.
Use a VM when you need actual "virtual" or "real" hardware access.

There is debate on nesting.  You can nest VMs/Containers etc.  within some limitations.  I for instance have LXCs which run docker.   VMs will run LXCs and dockers.  It tends not to work so well the other way up.

Where a VM is still better:
In my setup I have several "micro-VMs".  Little 1Gb or 2Gb or RAM things.  In several cases this was choosen because of simple things like USB.  While "dev files" map nicely into docker and LXCs some USB devices are a bit more awkward.  Things like multi-channel USB devices which don't present their endpoints until you enable them, then they are dynamically named by the driver.  These are a nightmare.  Not least the dynamic naming, which you can get around, but that the device proxying doesn't fully pass the mapped privelges, so you get "Permission Denined" on access.  In the end on these systems I went back to a VM.

The other place I used VMs was in the "storage" layer.  Containers, if they are to use any real userId or groupId have to be mapped via a file in /etc.  Like all container uids start at 10,000 or something and you need to map VuserIds to realuserIds.  For things like "NFS" which uses "client side id" verbatim this causes many problems that I didn't want to solve.  So NFS server got changed to a VM with real user Ids.

Also, almost all modern OSes support VMs out of the box without VMWare or other software.  Linux the subsystem is called KVM.  The kernel virtual machine.  I believe qemu is a front end for KVM.  Proxmox also.  Check the opensource projects for one that fits what you need.

For Windows it's "Enable HyperV" in settings and run HyperVManager.  For "docker" it's a bit more complex.  You can use plain old docker but it has issues in places.  "Docker desktop" which is a commercial "with community edition" application but supports more features.  Or... you can isntall "WSL" and just install docker there.  Again there are privelege limitations.

[Njk]

On the other hand, my DFU driver, which was signed with MS at the Win XP times, still work under Win 11. The 64-bit compilation anyway. And all the books about WDM and WDF driver development remains actual. In the economy class, the situation may be very different

[Karel]

Microsoft exec lashes out at AI skeptics, calls criticism "mind-blowing"

Stop being so ungrateful, people!

https://www.techspot.com/news/110330-microsoft-exec-lashes-out-ai-skeptics-calls-criticism.html

[DimitriP]

Microsoft exec lashes out at AI skeptics, calls criticism "mind-blowing"

Stop being so ungrateful, people!

https://www.techspot.com/news/110330-microsoft-exec-lashes-out-ai-skeptics-calls-criticism.html

Lets see for how much longer he stays there before "encouraged" to "explore other opportunities"  ....

[artag]

Microsoft says Windows is becoming an agentic OS, but users simply hate the idea

https://www.techspot.com/news/110225-microsoft-windows-becoming-agentic-os-but-users-simply.html

So pardon my ignorance, but I must be behind the times here:
Does "agentic" mean ""AI"-centered"? That term isn't familiar to me at all.

It's a marketing buzzword. It doesn't mean anything at all. What Microsoft's press release means is that they've latched onto a recent buzzword and are adding something into their dog's breakfast of an operating system that they can claim is new.
You can safely ignore it.
 

[SiliconWizard]

Microsoft exec lashes out at AI skeptics, calls criticism "mind-blowing"

Stop being so ungrateful, people!

https://www.techspot.com/news/110330-microsoft-exec-lashes-out-ai-skeptics-calls-criticism.html

Oh yeah, you're just a bunch of "AI deniers".

[Infraviolet]

I would say Windows is becoming a wonderful advertising campaign... for Linux

[Karel]

Windows updates keep breaking, and Microsoft's "agentic OS" isn't helping

Microsoft's latest patches risk turning the full Windows 11 shell into a non-functional disaster for enterprise customers

https://www.techspot.com/news/110498-windows-updates-keep-breaking-microsoft-agentic-os-isnt.html

[madires]

Microsoft seems to become better and better in screwing up updates. Could be a new business opportunity: Pay us or we'll update your OS!

[Karel]

Dell's CES 2026 chat was the most pleasingly un-AI briefing I've had in maybe 5 years

"One thing you'll notice is the message we delivered around our products was not AI-first," Dell head of product,
Kevin Terwilliger says with a smile. "So, a bit of a shift from a year ago where we were all about the AI PC."

It's not that Dell doesn't care about AI or AI PCs anymore, it's just that over the past year or so it's come to
realise that the consumer doesn't.

... what we've learned over the course of this year, especially from a consumer perspective, is they're not
buying based on AI," Terwilliger says bluntly. "In fact I think AI probably confuses them more than it helps
them understand a specific outcome."

https://www.pcgamer.com/hardware/dells-ces-2026-chat-was-the-most-pleasingly-un-ai-briefing-ive-had-in-maybe-5-years/

[Karel]

Conversations that Convert: Copilot Checkout and Brand Agents

Wherever shoppers are, they now expect AI to know what they need and help them act on it.

https://about.ads.microsoft.com/en/blog/post/january-2026/conversations-that-convert-copilot-checkout-and-brand-agents

Please, trust us with your creditcard!

[DimitriP]

Conversations that Convert: Copilot Checkout and Brand Agents

Wherever shoppers are, they now expect AI to know what they need and help them act on it.

https://about.ads.microsoft.com/en/blog/post/january-2026/conversations-that-convert-copilot-checkout-and-brand-agents

Please, trust us with your creditcard!

Oh No!!!.not an abandoned cart!!!

-"You viewed the page, now you have to buy something. We are not keeping the e-store up for just browsing".
-"But ..I use a "browser", not a "buyer". 

Maybe microsoft's next html client will be the Edge Buyer !
Which will put microsoft back to 19xx ...when they didn't have a browser