Omarchy: the Linux operating system that lets an AI agent redesign your desktop

🕒 Published on Zendoric: September 12, 2026 · 13:31
✨ AI-generated · how it's made
Nate opens his newsletter with an analogy: for AI agents, "it's 1997." They are showing up on forums, making requests over the internet, acquiring identities and exchanging information among themselves, much as humans explored the internet in its early years.
By Nate (Nate's Substack).
Nate opens his newsletter with an analogy: for AI agents, "it's 1997." They are showing up on forums, making requests across the internet, acquiring identities and exchanging information among themselves, much as humans explored the internet in its early years. As a concrete data point, he notes that on a forum built specifically for agents, twelve days of activity produced 20,040 posts and 192,410 comments. According to Nate, some of that content is useful and some is "ridiculous," in line with what happens when people experiment with something new before a mature, settled version exists.
The article's central argument is that this growing population of agents will soon need somewhere to work: compute, files, tools, memory and ways to communicate, including environments where they can keep operating after the person who assigned them a task closes the laptop. As agents do more, the design of those environments becomes important. In parallel, Nate notes that agents themselves are changing how easily the interfaces humans use can be built: much of what once required a product team and a roadmap can now start simply with someone describing what they want.
Both trends converge, the author says, in Omarchy, a Linux desktop assembled by David Heinemeier Hansson, creator of Ruby on Rails. Omarchy offers a complete and deliberately designed computing environment, and makes agents part of the process of modifying that environment: on first boot, the user picks an agent, and the project provides tools and instructions for asking that agent to help shape the system. Omarchy's own introduction describes itself, literally, as "the malleable OS for the age of agents."
Nate offers his own estimate — explicitly noting it is a personal expectation and not a data-backed market forecast — that around 1% of people will want to live at the flexibility extreme Omarchy represents, while the remaining 99% will keep using more predictable operating systems. Even so, he argues Omarchy can matter enormously even if that expectation holds. He says he himself will keep opening a Mac or a Windows laptop every morning, but that what is about to change is what his agents open.
The email also includes the table of contents for the topics covered in the full edition, reserved for paying subscribers (the text itself states: "Subscribers get the full deep-dive and guide, plus membership to my Slack community"):
- Why the audience for a software change can shrink to a single person, illustrated by the case of a father who built two different screen-time control systems for his two children, neither of which would be a product anyone would launch commercially.
- "The Omarchy curve": how much flexibility you get relative to the effort you are willing to invest, and why that curve is "bending" (that is, changing shape).
- What the other 99% of users will experience: Apple's September announcement is offered as an example of how most people will receive these advances, in the form of convenience built into a product they already know, rather than as an open, malleable system.
- Why agents are going to need their own computers: he notes that Cloudflare is already selling exactly that, and points out that desktop market share among humans offers no way to predict how many agent-oriented machines there will eventually be.
- A practical guide to what to change on an existing Mac or PC, which according to the contents includes specific changes to the machine itself, "the five tiers of access" an agent can request, and three different ways to try Omarchy itself.
The body of the email as received is the introduction/teaser for the full piece, so it does not develop these sections in detail; it limits itself to posing the question that closes the introduction: what happens when changing the decisions built into a computer becomes cheap enough to be ordinary, and when a growing share of the entities using computers — the agents themselves — can make those changes on their own.
🔗 Related on Zendoric


