Prompting My Mac
Last week, I decided my Dock icons were too large. Instead of navigating through System Settings, I ran a single command: nix-ask "make my dock icons smaller".
Within thirty seconds, the change was live. Claude Code had opened my Nix configuration, identified the Darwin module, adjusted the tilesize from 96 to 48, and rebuilt the system. This works because when a machine is defined entirely in code, an AI doesn't have to guess where settings are hidden; it simply reads the map.
Most macOS setups are a fragmented collection of .zshrc files, JSON snippets, and manual UI toggles. For an AI agent, this is a maze. In contrast, a Nix flake provides a structured hierarchy that is easy for a language model to parse:
flake.nix
├── modules/darwin/system.nix # OS-level: Dock, Finder, Keyboard
├── home/packages.nix # CLI & GUI applications
└── home/programs/neovim.nix # Editor-specific configuration
This structure allows for high-level requests that touch multiple layers of the OS. When I wanted to begin a project in Go, I didn't manually install binaries. I told the AI I needed a Go development environment. It added the compiler to my package list and configured the gopls language server within my Neovim module. I was writing code in a fully-equipped environment a minute later.
The same principle applies to aesthetics. Changing a system theme is usually tedious because it requires syncing the terminal, the editor, and the system UI. I recently asked for the "Catppuccin Mocha" theme across my entire stack. The AI updated my terminal colors, fzf variables, and Neovim theme simultaneously. Because the entire configuration is versioned in Git and managed by Nix, the risk is negligible. If I dislike a change, a git checkout or a Nix rollback returns the system to its previous state instantly.
The goal isn't necessarily to be "lazy" with configuration, but to eliminate the friction of system maintenance. I use a simple alias to bridge the gap:
nix-ask() {
(cd ~/nix-config && claude --dangerously-skip-permissions "$*")
}
Nix is known for its steep learning curve, but AI effectively flattens it. You don't need to be a Nix expert to maintain a perfectly managed machine; you just need a configuration structure that an AI can reason about. My Mac has shifted from a collection of settings I must manage to a system I can simply talk to.