How AI is changing the world - starting with mine

I’m a software architect by trade - a tinkering engineer at heart.

When it comes to LLMs and AI, it’s been a fast and wild ride - and I believe that we’re just at the beginning. I think there’s a pretty good chance that we’re going to get just another year or so down the road, and when looking back at today, it’s going to be difficult to reconcile how much this technology has advanced, how different much of the world’s day to day looks, and exactly how little time has actually passed.

There’s a lot of talk about bubbles, and maybe those concerns are valid - but I believe that even if they prove to be, we’re still at the beginning, and we’ve only just begun to realize some of the many, many ways that AI is going to change the world.

What I am capable of today, augmented with the latest and greatest in frontier models, harnesses, and orchestration tooling, far exceeds what I was capable of doing with AI just ~6 months ago.

And what’s even more interesting to me as a tinkerer is exactly how much advancement has been seen this year in the local / open source LLM space - with models capable of running on consumer hardware having matured from “an okay scripting assistant” to “an autonomous coding agent capable of delivering a non-trivial quality implementation based on a provided PRD in 1 turn”.

We’re not just talking about a small, but noticeable difference - we’re talking magnitudes, with each new month’s advancements dwarfing the 3 months that preceded.

As a software architect and engineer, of course, I use AI to augment my capabilities of designing and developing software. Honestly, I don’t know how engineers who haven’t embraced modern AI are staying afloat, as AI has accelerated seemingly every nook and cranny of our field, with even vocal skeptics like Linus Torvalds now embracing AI-assisted coding.

But aside from creating software, AI has completely changed how I use a computer - which is interesting to me, as I’ve largely been using a computer the same way for over 30 years up until the past ~2 years or so.

For example - hobby and home lab projects… As a tinkerer at heart, I’ve had no shortage of projects on my todo list since we got our first PC in about 1992…

Gone are the days where I sit down to setup some boring, yet needed, project in my home lab such as a multi-device localized backup solution, only to have the project that I planned on knocking out in an afternoon wind up spanning multiple days - as I find myself frantically Googling errors after the project inevitably wandered off of the happy path as a result of local environment quirks, or from following out-of-date documentation that’s still being served up on the project’s main docs page but has unknown to me been largely surpassed by the project’s community forum.

Now that flow is more like this:

mkdir ~/w/homelab/ez-sync/
cd ~/w/homelab/ez-sync/
touch ez-sync-project.md
code .
claude --dangerously-skip-permissions

I create a new home for my project, I create a file describing what I’m looking to accomplish, I open up the new project in VSCode, and I kick off Claude to help me flesh out the design for my project (often including Claude researching the latest and greatest in whatever space we’re working in to help define the solution), and then I interactively work with Claude to complete the project.

More often than not, the project still winds up taking a different amount of time than I initially anticipated - except now, it’s often less time rather than more…

My favorite approach isn’t to just let Claude run rogue on my system, but instead to assist in building out automation scripts and documentation that I can use to get things going more quickly. Artifacts that I can house in a private repo on either GitHub or my private Proxmox-hosted Gitea server, depending on sensitivity - so that I can reproduce the outcome predictably on command, rather than the ‘sauce’ vanishing with the Claude Code session.

Like many engineers, I have no shortage of backlogged projects: things I’ve identified a personal need for, projects online that piqued my interest, ideas I wanted to pursue but could never find the time for. While time is still very much a limiting factor, it’s not as obstructing as it once was - now, I can typically tackle that Saturday afternoon project knowing that odds are in my favor that I will actually complete it rather than it spanning multiple days at risk of abandonment.

This blog that you’re currently reading, for example - literally two decades of exporting and importing WordPress backups to new WordPress installs. Wrestling with getting all of the various dependencies, paths, and everything else required to host WordPress on my VPS - to ultimately walk away with something that was never quite me, and never quite what I wanted… But nowadays? A few hours with Claude describing what I want, how I want it to look, and what source material we’re working with (e.g. WordPress backups) - BOOM! I walk away with a custom Astro markdown powered solution that I can deploy on push, hosted for free on Cloudflare.

While I’m typing this, I have Claude and Codex debating the best way to modify my custom CMS powering The New Guard so that it can most easily support HTML or JSX files alongside the currently Markdown-driven content approach - which I should be able to get deployed before hitting the gym at 5… Something that previously would have likely taken me a few days of squeezing in a little work here and there between the more pressing demands of my day to day (family, pets, friends, job, cooking, cleaning, etc.), and so having a much higher risk of delay or abandonment, is now something I can knock out in 30-60 minutes while multitasking on other things…

We’re living in the future! I’m excited to see where this all takes us - and I’m really hoping that it’s not some dystopian hellscape, though I guess I’m going to enjoy the ride as long as I can regardless…

Have fun out there!

-Matt