diskiller's domain
← blog

Who Owns Your Computer?

2026-08-26

In February 2000 I wrote the following on this website, and thanks to the historical archive I can no longer pretend otherwise:

Unix as a Workstation? Well, its my honest opinion that this will never succeed. Linux won't make the desktop, nobody will.

Twenty-six years later I am writing this on a Mac, which is to say on a Unix workstation, and I am seriously considering replacing it with a Framework laptop running Linux. Governments are migrating public sector desktops to Linux and open source for reasons they describe as digital sovereignty. So: wrong, comprehensively.

But the interesting part is not that I got the prediction wrong. It is that I got the question wrong. In 2000 everyone argued about which operating system would win the desktop, as though the desktop were a trophy. That was never the fight worth having. The fight worth having is this one:

If the manufacturer holds privileges over your computer that you can never obtain, in what sense is it your computer?

This is not an Apple rant

I want to be clear about my position, because the argument gets dismissed the moment it looks like tribalism. I like Macs. The hardware is excellent. macOS is a genuinely good Unix workstation, and it has been my daily driver for years. My machines run Sequoia, which I consider the last good macOS release. I have no intention of installing Tahoe, whose interface I think is a significant regression, and which I am not able to decline in part while accepting the rest, because that has never been on the menu.

Liking Apple's products and rejecting Apple's model of ownership are not inconsistent positions. Apple is simply the clearest example, because it executes the model most competently and most completely. The argument is stronger coming from someone who uses the platform than from someone who never did.

The principle

The version of this argument you usually hear is "root should be root", which is too small. Here is the stronger form:

The manufacturer should not possess an irrevocable technical authority over a general-purpose computer that the owner cannot supersede.

Note what that does not say. It does not say platform security is bad. Nearly every mechanism people object to is, in isolation, good engineering:

The pattern is the same every time. The mechanism is fine. What is not fine is the mechanism terminating in a key I can never hold. And if I decide the operating system on my machine is bad, I should be able to erase it and install another one. On an iPhone I cannot, which is why I do not use one, and why I keep going back to Android in spite of its considerable flaws.

The false dichotomy

There is a real tension here and I am not going to wave it away. Apple's restrictive model delivers genuine security benefits to an enormous number of people who will never read a threat model. An iPhone is remarkably difficult to persistently compromise, and that is substantially because Apple controls the entire trust chain and refuses to share it. That outcome is real and it matters.

But the choice being presented is: 

security OR owner control

and that is a false dichotomy. We know how to build cryptographic trust systems whose root of trust is the owner. Enrollable keys, owner-signed images, documented and attestable transitions out of the vendor's chain of trust: none of this is unsolved computer science. It is solved, deployed, and boring. Allowing it costs the vendor support complexity and costs the vendor platform control. Those are business considerations. They are not physics, and they should not be presented as such.

The reasonable default is fine. Ship the locked, verified, vendor-signed configuration to everyone, because that is right for almost everyone. Just make sure a documented door exists for the owner who wants out, and make sure walking through it is a decision the owner can make rather than a decision the vendor makes on their behalf forever.

Where your data lives

The same principle applies one layer up, to services. My files sync between my machines with Syncthing, on storage I own. Synchronization is not what grants me access to my data. If Syncthing vanished tomorrow, every file would still be an ordinary file sitting in an ordinary directory on a filesystem I control, and I would be mildly inconvenienced while I replaced one moving part.

Compare the failure mode of building your digital life inside a vendor account. There, the service is not transporting your data, it is mediating your access to it. Lose the account, lose the platform, lose the subscription, and the question of whether you can still open your own photographs becomes a support ticket. This is why I barely use Google Drive either. It is not that these services are bad at their jobs. It is that I would rather my access to my own data not depend on anyone's continued goodwill.

The softer version: choice

User interface control is a milder instance of the same disagreement. Apple's historical attitude is unusually opinionated: this is how the product works. Sometimes that produces superb design, precisely because they refuse to accumulate twenty years of compatibility cruft and configuration switches. Sometimes it produces a change you consider a mistake, and the response to "give me the old behaviour back" is no, with a smile.

The pattern I object to is not opinionated defaults. It is removing choices rather than maintaining alternatives.

Linux is comically the opposite. Don't like GNOME? KDE. Don't like KDE? Cinnamon, XFCE, MATE, Sway, Hyprland. Replace the compositor, the shell, the display manager, the init system if you are feeling brave. Make your desktop look like Windows 7, or classic Mac OS, or a 1993 Unix workstation, or the bridge of the Enterprise. Occasionally the result is an incoherent mess. But it is your incoherent mess, which is the entire point.

Why this is bigger than laptops

Institutions are arriving at the same conclusion for the same reason, at a much larger scale. When a government migrates its desktops and infrastructure to open source, the motivation is rarely licence cost. It is that a public institution cannot accept a situation where a foreign commercial entity holds unilateral technical authority over the systems that run the state. That is the institutional form of the argument in this post.

It is also why open standards matter even when the proprietary implementation is objectively better. A documented format or protocol means the data and the workflow outlive the vendor. An excellent proprietary implementation of a documented standard is a fine thing to buy. An excellent proprietary implementation of an undocumented one is a mortgage.

And it is why I want Linux and the BSDs to remain competitive rather than wanting one of them to win. A monoculture on my side of the argument would end up with the same structural problem I am complaining about, just with different logos.

So, will I switch?

Probably eventually. Not yet, and not as a performance. My Macs run Sequoia and they do what I want, and there is no virtue in an ideological migration that makes my work worse. Framework plus Linux is philosophically attractive because the relationship on offer is roughly: here are the specifications for the machine you bought, here are the replacement parts, here is how to open it, install whatever operating system you like, have fun. Rather than a product that continually treats owner modification as something to be constrained.

The switch happens on the day a macOS release crosses the line where the advantages of the Mac no longer outweigh the restrictions attached to it. I am watching Tahoe's successor with interest.

Thirty years of computing later, the principle I have actually landed on is smaller and duller than the operating system wars I used to write about:

If I bought a general-purpose computer, its root of trust should ultimately terminate with me, not with its manufacturer.

Everything else is preference. That part is not.

© diskiller 1998-2026