Using AI for repetitive system tasks

Hello reader, welcome to my latest blog post!

In this one, I will chat a bit about my learning journey with my AI LLM of choice (currently Claude 3.7 Sonnet) and share a useful piece of software that I developed in just a few hours with its help. As I’m learning about this new technology, both the limitations of using chat for software development and the workarounds to address them are becoming clear. So far, it’s been a lot of fun to experiment with this new tech. After a relatively short period of adjustment to the norms of working in this paradigm, I’m becoming more prolific in generating useful code than I’ve ever been. In the past month, I’ve been able to use Claude to help me build my own LLM from scratch on local hardware, convert my existing self hosted container stacks to more modern docker compose stacks, and bring a new stack online for self hosting smart home services. Gone are the manual container orchestration and highly structured scripts. No more controlling disparate containers. I couldn't be happier. My new home labs are modern container stacks running with docker compose. They have monitoring, alerting, auto-healing, and auto-updating built in. My side project is coming along at triple the speed it did when I was hand coding. Things are pretty great, but there’s always room for improvement and I’m enjoying the process of improving my use of this tool a great deal.

“Hey, mister cloudy tech. That’s great that you’re having fun, but you mentioned free software?”

I sure did! It’s in my public GitHub, right here!

One of my core values in business and technology is the translation of “Nerd to English,” or converting tech jargon into plain speech. To that end, here’s my version of the half page story you wish didn’t precede the recipe. I posted it after the link to the “recipe” because honestly, if there’s anything the WHOLE world can agree on - it’s that we don’t need to know why this combination of ingredients is special to you. We just want a taste of the finished dish. I mean, it’s not like we’re debating the pronunciation of image formats, or some other massive concern that affects every person in the world /s. I’ll assume the technical folks are already in the repo and figuring out if or how they can use this thing, though I will vainly hope that they circle back to hear my story.

So what’s the big deal? It must be something, if I spent two whole paragraphs talking it up, right? The truth is that this script exists because, like a good engineer, I’m LAZY. Not (always) in the “Hey do you have a paperclip? Yeah, there’s been one of the carpet next to my desk for 6 months, use that one” kind of way. Rather, I’m lazy in the “Ugh, I need to do these annoying manual steps again?” kind of way. In this case, I’m talking about Windows 10 updates. I have no patience for the re-enabled auto-opt-in to usage reporting. I have less patience for one drive showing up. Again. Repeatedly. “Less than zero?!” you say? “Correct. Negative patience is what I have for One Drive.” I don’t want web search in my taskbar, I don’t want it pinned to edge, and I don’t want to disable it again every time I reinstall Windows. Edge is far from my preferred browser, but it's deeply embedded in the Windows ecosystem. It can't be fully removed and consistently tries to claw its way back to default status. I’ve never wanted Cortana. Once. Ever. But every time Windows update finishes and I reboot to the blue screen of setup, I have to tell it, AGAIN, that I don’t want any of the things Microsoft thinks I need. So I turn them all off, manually, only to find that one drive is signed in again and starting on boot. It's like 45 minutes unwinding stuff I already did months ago. Painful. It makes me wish Valve would release SteamOS for desktops. Then I could jettison Windows on the only system where I still 'need it' - my gaming rig. For what it’s worth, I think that the “You can game on Linux just fine these days” crowd is right. You can, and it works very well. I might be justly deserving of their righteous ire - IF I weren’t gaming on an HDR OLED TV. But I am, and support for such a system is a pain that I don’t want to take on, or continually/actively manage going forward, for my gaming rig. Between robust HDR support, display scaling, and QoL features, Windows sadly still wins on my TV connected gaming desktop system.

”But I have the tools to rebuild this. I have the technology. I can make it better than it was. Better, stronger, faster.” I whisper to myself. So Claude and I went to work. After all, with Windows 10 EOL only 6 months away, I’ll probably need to do this at least 5 more times. And I bet I can make it so I never have to do this again. And I bet it will take less time than it normally takes me to do this once. Spoiler alert, I was half right. It worked, but it took 4 hours. That said, overall, I’m quite happy with it.

So what is it?

Foremost, it’s a convenience script for folks with the same gripes I have. Digging deeper, it’s a modular system management script using only native powershell commands and no external dependencies. In other words, you don’t need to install anything else to use it. It exists exclusively to configure the system and unwind configurations that Microsoft re-enables after every update. It also handles those same things on a fresh install. It is designed with DevOps best practices in mind; it fails forward, it logs verbosely, and it’s highly configurable through command line options. Do you like One Drive, but the update you just did made edge your default browser again? Run it with the command line arguments to specify the default browser. It will intelligently determine which browser is installed and make that one the default without touching One Drive. Speaking of browsers, if it runs and it sees you have the two supported browsers installed, it tells you that it doesn’t know what you expect and does nothing. Then it provides you with the commands you would need to run in order to set the right default browser. It defaults to dry-run mode, in which it only outputs changes it WOULD have made if you’d given it the —apply argument. It’s configurable inside the script to always apply without the —apply option, should you be so brave as to run it automatically with no input as an administrator. It has a great amount of clear user feedback, including explaining steps that must be taken manually by a person (configuring the default browser, for example, can only be done by mouse clicks in the windows settings app).

And what is it NOT?
It’s not an enterprise system configuration management tool. It’s not something to add to your system startup that runs on boot or login. It’s not designed for unattended use.

But.. COULD it be any of those things?
Kind of. It’s never going to be a full scale enterprise configuration management tool. But it’s a great baseline for your own and the license attached allows you to extend it to your heart’s content - so long as you attribute the original author (me) and don’t charge for the script itself or the modification you made to it, it’s free for your use. Could it be something you add to your startup or be refactored for unattended use? Absolutely. It kind of already is these things, I just don’t recommend that anyone blindly use it as such. If you have the ability to work out how to use it unattended in a small business or home environment - you should! And I want to know how it goes, and if I can help you with your adaptation or use case. If you don’t know how to do that kind of thing… That’s what a Clouditect is for! Reach out so we can collaborate to make this or something like it work for you.