It feels like the entire internet has turned into a steaming heap of shit. Every major webpage has a multi-megabyte initial data transfer, then tack on JS bundles, one hundred thousand react useEffects, and 40 GraphQL queries. Everywhere I look, slop, slop, and more slop. I suppose it's where the term "enshittification" has come from.
motherfuckingwebsite.com explains it well. A good website is one that is lightweight, loads fast, fits on any screen size, looks the same in any browser, and is readable. This problem has existed long before generative AI, but it seems like since the era of "agentic coding" this has only become worse.
Perhaps this is because generative AI has given literally anybody the ability to use npm (inb4 "this should be bun, bro") to install a shedload of supply-chain compromised dependencies, write some typescript, tack on tailwindcss, and call it a day. You end up with something generic, heavy, and slow.
The transfer size of this post (as of writing) is 18.3kB. It honestly could be better. I built this website out in mid-2025 as a revamp of my much older design from 2021. The colour scheme hasn't changed, but everything else has. The stack is lightweight, just vanilla HTML/CSS/JS/PHP. The entire thing runs inside of a PHP docker container and is consuming 21.4M of memory. There's no postgres, no redis, nothing.
Humble brag over, the point I'm getting at is that this website works with a fraction of the resource usage of every vibe-coded AI slop website out there. This is just something that happens with AI generated websites, Garry Tan comes to mind, where he's saying "I'm being so productive, my AI is generating 30k LOC/day", and it's like, is that really productive? It just sounds like a lot of garbage in and garbage out to me. His shitty ruby-on-rails blog website returned an HTTP 503 when I tried to load it right now, after hanging for about 20 seconds. I assume one of his microservices has broken, again.
Lines of code is not. an. indicator. of. quality. It is a facade vibe coders like to use as a way of pretending that they're getting things done. "My Openclaw Agent made 300 commits today", that's awesome, are any of them usable? This sort of grandstanding ends up creating a massive maintenance liability. If these are "contributions" to open source (which many of these performative posters focus on), then a human will eventually have to read, review, and maintain this behemoth of code. More often than not, it's horribly inefficient, ten times as long as it needs to be, and works only in the exact specific scenario it has been written for.
---
Don't get me wrong - I use AI. I like AI. Just earlier today I set it to the task of reverse engineering Gigabyte's awful motherboard UEFI firmware to troubleshoot an issue with S3 sleep. Agentic AI can speed these sorts of tasks, which might take me a couple hours, up significantly. Need some research done? It can search the web, compile information, and produce something of genuine value. But for me, this is really where the "magic" of AI stops, because it is just not capable of creating something new. This is why every vibe-coded website feels the same, it's why some of the first things it'll suggest you to build when you ask it is a note taking app, or a reminder app. It's why every single AI B2B SaaS does the exact same thing. It's boring.
Large langage models are inherently backwards-facing, they can only reference the datasets that they have been trained on. They simply can not come up with something completely new. If it doesn't exist in the model weights, it cannot output it. Creativity as it's own concept is forward-facing, it's the creation of new ideas, things that haven't been thought of before. One could argue that LLMs can generate some interpolation of ideas that might seem new, but there's a complete lack of intent behind it. It's not creativity, it's throwing darts at a board to see what sticks (as I wrote this, I felt an odd uncanny valley sense from GPT-4o's "it's not x, it's y").