I Was an AI Skeptic. Here's What Changed My Mind.

I'll be honest — I didn't get the AI hype for the longest time.

I had a paid ChatGPT subscription and barely used it. I'd ask a question, get an answer, close the tab. A slightly fancier search engine. After a couple months of that, I cancelled it. I wasn't getting enough out of it to justify paying.

Being the homelab person I am, I self-hosted OpenWebUI — figured if I was going to use AI at all, I'd rather keep it local and private. Same story though. I'd poke at it occasionally, but I still couldn't find a real use case that stuck.

What changed everything was context.

I run a homelab — a couple of machines at home running self-hosted services, media automation, Docker containers, the whole deal. When I started using Claude Code and it was sitting right there in my terminal with access to my actual code, my conventions, my project structure — that's when things got interesting.

Tweaks I'd always wanted to make but never had the time for became possible. I wasn't starting from scratch anymore. I had a blueprint. My job became reviewing, testing, and tweaking instead of writing everything line by line.

The first real "okay, this is different" moment was when I forked an open source project called Booklore. It's a self-hosted book management tool, and there were bugs that had been bothering me for months — syncing to Hardcover wasn't working properly, ratings weren't pushing through, and there were security vulnerabilities sitting there untouched. The original maintainer just wasn't focused on fixing them.

Without Claude Code, I never would have attempted it. Looking at someone else's unfamiliar codebase with my schedule? That's a months-long project I'd never start. It became a weekend project. I fixed the sync, got ratings and shelf management working properly, patched the security issues. It all just works now.

But the thing that really sold me wasn't even the code stuff — it was Claude's Projects feature.

You can create separate projects, each with their own context and instructions. So I have one for my homelab where it knows my entire setup. I have another for health and fitness where I uploaded my workout history, my nutrition logs, and my goals.

That fitness project is where it clicked on a completely different level. It looked at my data and found a pattern I couldn't see myself — I have a tendency to go all in too hard, too fast, then completely burn out for weeks, negating any progress I'd made. Through conversation, it figured out that I wasn't getting enough fiber, which was throwing off my hunger cues and leading to binge eating. For workouts, it looked at my calendar and found realistic time slots instead of an ambitious schedule I'd never stick to.

Now it gives me a meal prep schedule that works around my actual life — I told it I don't like cooking big meals in the middle of the week and I need variety or I get bored. When I'm at the grocery store, I take a picture of a product and it tells me whether it fits my goals or if I should put it back.

Then I went deeper. I built an MCP server — basically a way for Claude to talk directly to my self-hosted services through chat, even when I'm not at my terminal. I connected it to my media tracking, my reading lists, my audiobook library, my homelab.

Now I can ask what I should watch next and the recommendations actually feel personal because it knows what I've already seen and what I'm into. I can ask about a book I'm reading and it won't spoil anything because it knows exactly where I am in it. I can add something to my watchlist just by asking. I can ask questions about my homelab setup without pulling up a terminal. It all just works through a conversation.

And the thing is, it doesn't stop. Every time I think I have a full picture of what's possible, a new use case pops up. I don't sit on ideas anymore wondering if something is feasible and then never doing anything about it. I just try it. It's scratched an itch in me that I think every engineer has — that urge to experiment and build, but without the barrier of needing weeks of free time to pull it off.

I went from cancelling my ChatGPT subscription to building Claude into nearly every part of my workflow. And I don't think I'm done finding uses for it yet.