Tailwind and slop apps

(In my experience, it's never been "this doesn't look as good as the latest version of Discord", or whatever.)

pair that with the field of ui/ux thinking landing pages that say nothing at all and you get a million webpages that look the same. every sass does this. in fact, every saas had a slop feeling to it because of this before AI took off. before chat friggen jippity and now claude and gemini were ubiquitous, i was regularly rolling my eyes at every. single. saas. website.

all the examples in OP don't look like a tailwind specific thing to me at all. they look like a broader boring internet-webdev equivalent to live laugh love pinterest home interior designs that predates the AI slopfest. this is the web styling VC's have been demanding from the startups they've invested in for many years now.

as far as tailwind goes, it's more like a utility set of classes but i see no reason why you can't get creative with it. feel like it's catching strays here simply because everyone's vibe coding and it's a common path for trained models to choose but that really says not a lot about tailwind and a whole lot about how people are choosing to use tailwind. it's not like tailwinds providing a breadth of components or is locking in certain stylistic elements. it's just that everyone's making the same choices because everyone's slopping their front ends.

Bootstrap ships components. AFAIK you need another library if you want that in Tailwind.

yes, and who wants creativity and risk when everything can look like the interior of a McDonalds. I'd much rather look at someone's terrible and scuffed attempt at designing their own page, because it at least signals that there is a human who isn't afraid to try something out rather than the Instagram filter version of a webpage.

You can either please the Greeks or the Trojans, I guess.

A “perfect” website has things cleanly separated out between style, logic, and DOM, but that does cause cases where you might have to edit three files every time you change a single widget of your website. Neglecting to do that can leave dead or incorrect code just because the issues do not inherently present themselves to you.

Yes, a perfectly principled programmer would do it right… but have you worked with other people before? They’re not necessarily very reliable.

1. Locality of behavior - the classes are the description of the style once you know tailwind, and they live where the thing they're styling is. I can look at a template and know pretty well how it is going to look without having to go lookup a sites personal bespoke CSS and match what's written there to the markup.

2. It's the same classes as you move from project to project. I have tons of small one-off apps I maintain, all using tailwind and I don't have to figure out the custom CSS classes and their meanings. It is a knowledge set that reduces friction between projects and I really need things that reduce cognitive load like that.

3. No dead CSS. There are tools to help with this in regular CSS but I just don't have to think about it with tailwind and that's nice. I work on my templates and the CSS is updated and minifed on the fly without extra steps (or unnecessary CSS).

4. Easy to build a personal component library that doesn't also require relevant CSS to be bundles with it. For me this one is pretty great. Thanks to template engines (for me it's Django's, but Jinja or others are grand too) I have a simple place to pull components (really just template partials) and widgets I've made and they just work because the underlying classes are the same. There are some exceptions around color schemes and the like but now all my little components take a colour scheme arg and voila, no proballo.

I think the key here is that a description of the visual is embedded in the markup. That lets the actual mark-up delivered to the front-end describe pretty much the whole kit. With server-side rendering ( I use HTMX ) I can even write unittests that responses contain expected markup, including CSS classes (I built a few helper assertions that all of my test suites use) which gives me some interesting checks.

I really ultimately don't think there is one correct way to do things. The things I value may be utterly irrelevant to another developer.

The best way to figure out how to prompt an LLM is to develop design skills so you know what to tell it.

Same as you have to be explicit about instructions to humans.

Even humans use their default assumptions w/o clear directions

I think what they're actually experiencing is that they can instantly tell a default Tailwind site with reused components, which is not much of a claim at all.

the problem is not tailwind, the model output just reflects the training material that most likely overindexed on tailwind templates with lots of fonts and text sizes and outlines, I'm sure future training runs will be better

Everything else is just people using a default configuration instead of building their own.