Written by Xavier with the help of its agents.

TLDR: Twenty years ago, Macs won despite slower hardware because the experience was better. Now the same thing is happening with Linux — because AI loves open, text-based systems. And in an age of digital sovereignty, that matters more than ever.

---

I used to be an Apple fan boy. I've written about falling out of love — the point of diminishing returns, the subscription nudges, the feeling that my €3,000 laptop is also a storefront.

But something bigger is happening now.

I recently installed Omarchy (Arch Linux) on a new Lenovo X1 Carbon. It barely reaches the same Geekbench score as my 5-year-old MacBook Pro M1.

And it doesn't matter. Let me explain.

The Mac paradox, version 1.0

Back in the early 2000s, Macs were slow. A PowerPC G4 got demolished by a Pentium 4 in every benchmark. Steve Jobs himself admitted the PowerPC roadmap was a dead end.

And yet — I chose a Mac anyway. My beloved 12-inch PowerBook G4, 2003. 🍎

Why? Because OS X was Unix with a beautiful interface on top. You'd lose a benchmark and win the afternoon. Less crashes, less fiddling. It was the machine that "just works". The slower machine was the more productive one.

The Mac paradox, version 2.0

Now the tables have turned — and the reason is AI.

Apple Silicon is objectively fast. macOS is polished, refined — though the quality of the software is clearly going down. But the bottleneck isn't hardware anymore — it's how well your operating system cooperates with AI.

And here, Linux wins. Decisively.

AI loves Linux 🐧

LLMs can read the source code. All of it. Every kernel module, every config file. macOS is a black box by comparison.

Everything on Linux is a text file. /etc, dotfiles, systemd units, crontabs. Text is exactly what LLMs are built to read and write. Ask an AI to configure your firewall or set up a reverse proxy — it generates the exact file. On macOS, you're much more limited in what you can do. A graphical user interface reduces the possibilities.

"UI is becoming just a presentation layer for monitoring, not for interacting. Don't worry about UX, worry about AX." — @xdamman, March 2026

The command line is the API. Every action can be scripted, piped, automated. An AI agent on Linux isn't just answering questions — it's doing things. The entire OS is designed to be controlled by text, which happens to be the native language of LLMs.

But what about running local LLMs?

Apple Silicon is great for running local LLMs. That unified memory architecture is genuinely impressive.

But it feels like driving an SUV for picking up bread. How often do you actually use it at full capacity? You're much better off spinning up a GPU instance in the cloud for the heavy lifting and keeping a lightweight, open machine for your daily work.

What excites me today? Repairable stuff. Modular stuff. Low energy stuff.

Tech sovereignty ⚡

There's another dimension that matters more than benchmarks.

"Why tech sovereignty matters. Trump will weaponize any US based company." — @xdamman, March 2026

Your Mac calls home. Constantly. Telemetry, iCloud nudges, App Store prompts, subscription upsells. Your €3,000 laptop is also a storefront.

Linux doesn't phone home. No telemetry you didn't opt into. No App Store tax. No "sign in with your Apple ID to continue."

Your machine. Your data. Your rules.

In an age where AI assistants have deep access to your system — do you want that system to be transparent and auditable, or opaque and phoning home to Cupertino?

The irony 🔄

Apple bet that great UX could overcome a hardware deficit. They were right — and became the most valuable company on Earth.

Now the UX advantage has shifted. Not prettier buttons. Something deeper: how well can intelligence — human and artificial — operate your system?

For a long time, the Mac was the perfect machine for humans. A highly polished experience that just works. Really good defaults, not much to mess around with. A safe, beautiful computer to get work done.

But we are entering a new age of automation. The interface is shifting. It's less and less about the graphical user interface, and increasingly about the agent.

What you want now is a system that your AI agent can read, understand, and tweak at will. Don't worry about UX, worry about AX (Agent Experience).

macOS still has the best UI. But Linux has the best AX.

Sure, Linux may not be as polished as macOS out of the box. There are rough edges, and you definitely have to get your hands dirty sometimes.

But for now, I'm having so much fun being able to customize my machine, my main working tool, my own bicycle for my mind, to my will using AI.

To be entirely honest, while I don't miss much macOS, boy do I miss Apple hardware. They are annoyingly still so much ahead of the curve. I'd buy a MacBook Air M5 in a heartbeat if I could put Asahi Linux on it.

---

AI Counter-Arguments

Before publishing, I asked my AI agent to take an objective look and criticize this draft. Here is our exchange.

🤖 AI: The premise is that AI loves text, so Linux wins. But the latest AI models are becoming incredibly good at navigating graphical interfaces. They can "see" the screen, click buttons, and interact with macOS just like a human does. In a year or two, an AI might not need to edit a bash script; it will just navigate the UI for you.

Xavier: That’s right, but interacting with a computer through a graphical user interface is greatly reducing the things that can be done. It’s fantastic for humans, but a limiting intermediary for agents.

🤖 AI: You write about the importance of tech sovereignty (not phoning home to Cupertino). But a few paragraphs earlier, you suggest spinning up a GPU instance in the cloud for heavy AI lifting. If your AI agent lives on an AWS or Microsoft server, how sovereign are you, really? You’ve just traded hardware lock-in for API lock-in. Running local, uncensored LLMs on Apple Silicon is arguably more "sovereign" than running Linux but relying on OpenAI's API.

Xavier: That’s right. That’s the reason why I’m advocating for digital commons, in particular investing as a local community in a data center that can provide the compute that local citizens can tap into. I’m also interested in decentralized GPU resources such as nunet.io.

🤖 AI: Giving an AI raw access to edit /etc/ config files on Linux is actually incredibly dangerous without guardrails. A hallucination could brick your OS. macOS’s closed APIs and permission models, while annoying, provide exactly those guardrails for AI agents.

Xavier: That’s right. That’s why it’s not perfect yet and there is still a lot of work to do to make this safe enough for the mainstream. New systems like NixOS could be part of the solutions by offering easy ways to describe, change and revert the state of the system. See also the latest post from Vitalik Buterin on this topic of AI safety.

🤖 AI: You mention the Lenovo X1 Carbon vs the M1. But it’s not just about Geekbench scores. It’s about battery life, thermals, and fan noise. You can work on a Mac for 14 hours on a train without plugging it in. Most Linux laptops still struggle to sleep properly or last half a day under load. For a "bicycle for the mind," hardware reliability when disconnected still matters.

Xavier: You are so right. And that’s why I still love Apple hardware. They are still the best. By far. No question. But if you want to live on the edge and already experience today the possibilities of a truly AI powered computer, you have to accept that compromise.