I’ve spent years using Linux because I want control over my machine.
Then I installed something that basically said:
“Give the AI access. It’ll handle it.” 😅
So I did.
I let Omarchy’s AI modify my desktop, create things, troubleshoot the OS, and even deal with something I intentionally broke.
The interesting part wasn’t whether it could write commands.
We already know AI can do that.
The interesting part was watching AI actually operate the Linux desktop for me.
This feels like a preview of where Linux is heading.
Full experiment 👇
https://t.co/tHc9GLjIg6
Linux has always been about control.
You know what runs.
You know what changes.
And when something breaks, you usually know exactly who to blame.
Then I gave Omarchy’s AI access to my desktop.
I asked it to change my environment, build things, troubleshoot problems, and make decisions directly on the system.
That is where it got interesting.
Because once AI can actually touch your OS, this stops being “just another AI assistant.”
It starts feeling like a completely different way to use Linux.
So I tested it.
And yes, I intentionally broke things too. 😂
Full video 👇
https://t.co/tHc9GLjaqy
I Trusted Omarchy’s AI With My Linux Desktop
I gave Omarchy’s AI access to my Linux desktop.
Not a chatbot.
Not a terminal assistant.
It could actually change things on my machine.
So naturally, I trusted it with way more than I probably should have. 😂
Here’s what happened 👇
https://t.co/tHc9GLjIg6
I Trusted Omarchy’s AI With My Linux Desktop
@dhh The hard part was never having power.
The hard part is building systems, institutions, and people disciplined enough not to use it just because they can.
Everyone is suddenly obsessed with AI agents.
Meanwhile NixOS has been sitting there like:
“Cool. Want the agent’s entire environment to actually be reproducible too?”
That’s the part of NixOS I didn’t appreciate until I used it.
CTRL+Chaos👇
https://t.co/7an6gqfppL
Production engineering is 10% architecture, 20% YAML, 30% dashboards, and 40% “who changed this?” I made a channel about the middle part.
CTRL+CHAOS 👇
https://t.co/xszOtfS5Wq
If you believe production should be treated like a museum on Fridays…
Look. Observe. Appreciate.
But DO NOT TOUCH anything. 😂
You’ll probably enjoy my YouTube channel.
Linux, SRE, Kubernetes, cloud, and all the chaos in between.
CTRL+CHAOS 👇
https://t.co/xszOtfSDLY
CTRL+CHAOS is basically me poking production with a stick and asking:
“why are you like this?”
Linux, SRE, Kubernetes, AI infra, weird failures.
https://t.co/A81vRERhqo
@mattpocockuk This is the part people underrate.
A better model helps.
But an agent that can inspect the app, test its own changes, see failures, and iterate inside a real environment is a completely different thing.
The environment is part of the intelligence.
@theinformation Funny how fast “AI company” turns into “infrastructure company” once customers show up.
Models get the headlines. Capacity planning gets the pager.
@sama This is bigger than login.
If my AI subscription, context, and tools can follow me across apps, the model starts feeling less like a product and more like infrastructure.
That gets interesting fast.
NixOS is one of those ideas that sounds obviously right once you understand it.
Then you try to use it and realize the idea is the easy part.
That tension is basically the whole video.
https://t.co/7an6gqfppL
Imagine rebuilding your entire Linux workstation on another machine without remembering every package you installed or every setting you changed.
That’s what got me interested in NixOS.
The concept is fantastic. Actually getting comfortable with it is another matter.
https://t.co/dtgYjurm4g
NixOS has me questioning why we’re still configuring Linux machines by running 47 commands and hoping we remember what we changed six months later.
Then I spent an evening fighting my own Nix configuration.
https://t.co/7an6gqfppL
I think Omarchy has a bigger problem than people realize.
It makes me want to mess with my Linux desktop again instead of actually getting work done.
Thanks, @dhh. Very productive. 😎
My review:
https://t.co/gSV2JCDVhu
@SawyerMerritt@elonmusk I’d love to see more independent testing of FSD in actual European traffic. Different roads, different driving habits, different edge cases. That’s a much more interesting engineering discussion than arguing over approval dates.
@tobi I’d be more interested in seeing the same video generated with 10 completely different prompts. What the AI leaves out might tell us more than what it includes.
Linux spent decades making tiny commands boring and predictable.
Now Ubuntu is replacing GNU coreutils with Rust, and we’re finding out how much software depends on those boring little details.
Memory safety matters. So does not breaking stuff.
I tested it:
https://t.co/d2Yy2Gpjxk