wow, cool website. shame Claude made it.

You know, I've had medidew.name for a long time now. Around half a year, I think. I originally only bought it because I needed a domain to host a Bluesky PDS and a Matrix server, and it wasn't exactly a huge expense to put towards that purpose.

I also knew, once I made that purchase, that it would be a waste not to host a website of my own with it, completely independently, without relying on any web-hosting service. There's just something exciting about being able to say 'I own and run every part of my own website' that appeals to the same giddy part of me that got into self-hosting in the first place.

However, if the rant on the front page of this website didn't already spoil it for you, I really hate web design. It's not something I've ever had an interest in learning.

In fact, I hate most professional-looking websites. I spend most of my time in a terminal, so why would I care for fancy graphics or pretty colour schemes? I know they have a place, and I'm not trying to belittle the skillset of any graphic designers out there, but too many times have I visited a website looking for one specific piece of information only for it to be hidden in the pocket of some animation I can't search — because it technically isn't text, but instead a static image.

But most of all, it's just not me. If medidew.name is supposed to be my website, then I want it to reflect that. I know what I value in my user experience and that's functionality above all else. I'm not interested in catching everyone's eye or showing off. This is my turf and if I want to put a spinning rat in the corner of your screen, then I'll put a spinning rat in the corner of your screen.

Okay, but you still had to sit down and write the HTML templates.

That's the neat thing: I didn't!

The reason this website exists at all is because I very recently got a Claude Code subscription for the first time, and creating this website was my first time giving it a spin. While I've tried running LLM models off personal hardware before (because I like my privacy and like to get the most out of what I already have, thanks), there's only so much I can do with an RTX 2060 and a dream.

So I figured, if I don't want to deal with any of the front-end elements of a website, but I haven't been able to host any models powerful enough to write decent code on my own machines, why not tell Claude to do that for me? Hell, why not tell it to make the entire website and see what happens?

It's safe to say that it worked like a charm.

Up until now, my opinion on the legitimate applications of enterprise LLM models has been split. I've toyed around with 'em before, and they're certainly impressive, but I've never felt like I've been missing out by avoiding them. There has always been a very vocal group saying they are useless, expensive, and actually make life harder. There has been a less vocal group saying the opposite.

I don't have experience using LLMs in a professional work environment, so it's hard for me to form an educated opinion on the matter, and out of the people I know who do have that experience there still isn't a clear consensus on whether these models actually boost their productivity. Some people love them and some people hate them. Software engineering is probably the only field where I'd say I've heard more positive feedback than negative.

So, going into using Claude Code for the first time, I wasn't sure if I was going to have to spend eons debugging the simplest mistakes or if it was going to do what I asked perfectly on the very first try. Logic dictates that if it wasn't useful then people wouldn't spend money on it, but LLMs have also undergone a lot of hype, and sometimes markets make poor decisions. I wanted to know for myself.

Claude's pros have thus far outweighed the cons.

It remains to be seen how Claude handles more complex tasks; I plan on getting it to try and write a parser for Deadlock demos later, but my experience so far is leaning more towards the latter, 'perfect' experience. I'm not sure how I feel about that.

Part of me thinks it's really cool. I did not expect Claude to check my past projects for insight into my coding style, for example, but I was pleasantly surprised when it did. I went in half-expecting an error fest, but I've had nothing of the sort.

Part of me also thinks that's really boring. Code performance might be something you can evaluate objectively and optimise, but I still see it as an art form of sorts. There's nothing more satisfying than writing a really elegant solution to a problem, but if Claude can do it faster than I can then what's the point?

I really wanted to hate it.

I'll spare you my full thoughts on LLMs and Generative AI for another time, but let's just say that while I recognise the merits of the technology — especially in software engineering — that doesn't mean I want it to be stuffed into every facet of my life. I prefer to read books written by real people, to listen to music made with heart, and I will always tread around the technology lightly so long as the long-term repercussions of LLM exposure are poorly understood.

Chatbots can drive people to the point of psychosis, our critical thinking skills may be diminished by over-reliance on personal assistants, and most people do not understand how this technology truly works.

The fact is, LLMs are yet another technology that is fundamentally changing the way we consume information, and if the internet and social media have taught me anything, it's that not everything will change for the better. I think it's a mistake for companies to push this tech so hard for profit before knowing whether it's in the public's best interests to do so.

Also, I do not need AI in my bloody TV.

Just for once I wish we would take the time to stop and think before thrusting new technology into public hands, but that's idealism speaking. I know it is. But it's fine. Ethics is always playing catch-up, and Pandora's box is open now anyway. I begrudgingly accept that Claude Code is useful and I will continue to use it, but don't expect me to start using Opus for every little thing.

I'm excited to continue testing Claude Code's limits nonetheless. A little sad at how it's going to change my workflow maybe, but I'll adapt. At least now I have a neat little website and blog to post my rambles and projects on.