Made to Measure

In the late 1990s, I spent an unreasonable amount of time making my Linux desktop mine.

I have written before about those days: the community site, the boxes of CDs that distributions sent us, loving SuSE before settling on Debian. But what I remember most is not the distributions themselves.

It is the customizing.

Every setup was a little different. Window managers, dotfiles, hand-written scripts, prompts tuned to the pixel. You could lose an entire weekend getting your terminal to look and behave exactly the way you wanted. I know I’m a big nerd, but I had cool bash prompts before they were popular. And not to mention building every single new version of the Linux kernel by hand. I wonder how many hours I spent watching “make” run.

But… This story isn’t about nostalgia.

The point is that personalization used to be something you paid for with time. Lots of it.

But then life got busy; I had a real job, a family, and less free time to keep customizing stuff. While I have carried my home directory with me for maybe 20 years or so, it is mostly things I put together a long time ago. Recent additions were mostly small functions to quickly wrap other things, or little tweaks to tmux and whatnot.

Linux has advanced a lot, and customization has become less and less necessary. I can’t even remember the last time I built my own kernel. Maybe 15 years ago?

I just had to change the colorscheme, fonts and wallpaper and I was good to go in most cases. Practical, easy, quick. But a little less fun.

But then Omarchy caught my attention.

Omarchy is DHH’s Linux setup: Arch, the Hyprland window manager, a tightly curated set of tools. It is the environment he uses every day, packaged so that anyone can install it. It started as a post-install script and grew into a full distribution, with its own installer. It is aggressively opinionated: strong defaults, keyboard-driven, no framework of endless choices. You take his decisions, and you get on with your life. Or you change everything with AI.

DHH envisioned Omarchy as an AI-first Linux distro. It is properly populated with AGENTS.md and SKILLS files, so the system’s inner workings are clearly delineated for agents like Claude, Codex, or Grok. It is beautiful!

That lets someone tweak small (or large) parts of Omarchy without needing to know or care about syntax, configuration files, and whatnot.

This is huge. And then something came to mind. A friend of mine, who isn’t in technology and sometimes has trouble with basic computer stuff, got ChatGPT (via the web app, of all things) to build him a personalized Android application. Then it walked him through enabling developer mode and deploying the app himself. He wanted something very specific, tailored for a single user: himself. That wouldn’t have been possible until six to twelve months ago. Even we techies with software development backgrounds have graveyards of pet projects that never saw the light of day because coding by hand takes time we don’t have.

And that used to apply to companies too.

For most of my career, “custom software” meant a project. Scope, budget, vendor, timeline. The small, weird need: a report only your business cares about, an integration between two tools nobody else uses together, a dashboard that matches how one particular team actually works. They wouldn’t happen.

So people lived with the standard thing. Plus spreadsheets. Plus copy-paste.

In a previous job, we had an entire department dedicated to translating customer spreadsheets into a format the inventory system would understand. The customer was huge, and we were small, so they changed things frequently and didn’t give a crap if we needed to rewrite an entire parsing system on two days’ notice. Almost every month, something was utterly broken, and we wasted hours fixing it.

Again, it was a large customer, but not the only customer. But they needed a highly customized parser just for them. That was years ago.

Today, an AI agent builds that in fifteen minutes.

And I think that is the interesting part: AI is changing the economics of customization.

For a long time, we had to choose between adapting ourselves to the software or spending a lot of money and time making the software adapt to us. Most of the time, we chose the first option. It was simply cheaper.

That equation is changing.

If I can tell an AI agent, “I want this application to work this way,” and have it build the missing piece in minutes, there is much less reason to put up with software that almost does what I need. The same applies to my Linux desktop. I don’t need to learn the syntax of every configuration file anymore. I can explain what I want and let the machine figure out how to make it happen.

In a weird way, AI might bring us back to the old days of computing.

Not because we will all start compiling kernels again, but because computers can once again become deeply personal. The difference is that the barrier is no longer knowing how to write the code. The computer can help with that part.

The nerd in me finds this fascinating.

Back then, I spent entire weekends customizing my computer because I had to. Today, I can spend fifteen minutes describing what I want and have an AI agent do most of the work. And that doesn’t just mean prettier terminals or clever bash prompts. It means building the weird little tools that weren’t worth building before.

Maybe that’s the real promise of AI, lost in all the talk about replacing programmers and automating jobs.

It makes software cheap enough to be personal again.

And I think that’s pretty damn cool.

If you have a process that is too specific for off-the-shelf software, or a small internal tool that nobody has the time to build, I can help. I use AI, automation, and my 25+ years of experience in infrastructure and software engineering to turn those weird little ideas into useful things without turning them into six-month projects. Let’s talk.