> I got curious about what writing more semantic HTML would feel like.
I've been teaching semantic HTML / accessible markup for a long time, and have worked extensively on sites and apps designed for screen readers.
The biggest problem with Tailwind is that it inverts the order that you should be thinking about HTML and CSS.
HTML is marking up the meaning of the document. You should start there. Then style with CSS. If you need extra elements for styling at that point, you might use a div or span (but you should ask yourself if there's something better first).
Tailwind instead pushes the dev into a CSS-first approach. You think about the Tailwind classes you want, and then throw yet-another-div into the DOM just to have an element to hang your classes on.
Tailwind makes you worse as a web developer from a skill standpoint, since part of your skill should be to produce future-proof readable HTML and CSS that it usable by all users and generally matches the HTML and CSS specs. But devs haven't cared about that for years, so it makes sense that Tailwind got so popular. It solved the "I'm building React components" approach to HTML and CSS authoring and codified div soup as a desirable outcome.
Tailwind clearly never cared about any of this. The opening example on Tailwind's website is nothing but divs and spans. It's proven to be a terrible education for new developers, and has contributed to the div soup that LLMs will output unless nudged and begged to do otherwise.
Treating markup and styles separately is great, in principle, but you'll always need additional markup for certain things. We knew this going back to the early 2000s.
There is nothing about Tailwind itself that forces you to use divs and spans instead of the appropriate HTML tag.
Documents and interfaces are different. Tailwind makes a lot more sense for interfaces. You can use Tailwind for the interface and scoped HTML selectors for other content.
Tailwind is around 4x faster and has practically no overhead compared to writing a complex CSS codebase. Whatever you think of it, this is always a benefit in its corner.
While I agree I do think there's some "aspiration of purity/correctness" in your approach that I've long let go of.
I look at the royal mess that is HTML/CSS/JS as a necessary evil, required when we want to target browsers. To me it's "just the presentation layer".
In my work I put a lot more emphasis on correctness in the db schema, or business logic in the backend.
When it comes to the messy presentation layer I prefer to write a little as possible, while still ending up with somewhat maintainable code. And for this Tailwind fits the bill really well: LLMs write it very well, new devs understand it quick, and it's quite easy to read-back/adjust the code later.
I 100% agree a Tailwind project is not the best way for a new dev to learn HTML/CSS. But then I prefer the new dev to focus on great db schemas, intuitive APIs, test-able biz logic, etc. Fiddling with the mess that's HTML/CSS is not the place where I consider human attention is best spent on (or where developers pick up skills to become much better developers).
This isn't about "purity/correctness" it's about the real experience of a blind person. Accessibility means caring about the HTML.
Your comment only mentions developers as the audience of HTML authoring, as opposed to users, which is a common attitude and the core problem with Tailwind.
I use Tailwind and have all kinds of "screen reader" directives in my templates.
Not sure if it helps, but if we get our first blind user I will gladly make some admends to make it more usable for them.
It seems that Tailwind is now blamed for the mess that is HTML/CSS. Tailwind certainly allows for accessible designs; it may not be the ideal solution, sure, but what we aim for is "good enough".
> Tailwind instead pushes the dev into a CSS-first approach. You think about the Tailwind classes you want, and then throw yet-another-div into the DOM just to have an element to hang your classes on.
I wholeheartedly disagree. That mindset is not caused by Tailwind, but by being ignorant.
You can perfectly create an HTML document with semantic meaning and the add Tailwind just as any other CSS framework or pure CSS to it.
And DIVs do not carry meaning, they are specifically to add functionality or styling, so you can throw in as many as you like. Using them abundantly isn't good style, but the way you make it sound that they're evil isn't good either.
You're not wrong, and I mostly agree with you. I die inside when I see the div soup that a lot of sites have become. However, I think there is value in being able to have the important parts of CSS merged into the HTML a bit. Where that line is, is certainly up for debate (and I don't have the answer), but I've found a lot of my tailwind sites are more readable to me than my pre-tailwind sites, often because I don't have to context-switch and open a different file to be able to reason about the styling on an element. For big stuff the second file can be nice, but there's a lot of style tweaking that is great to be able to do right there in the HTML. Tailwind does really lead you to ignore the css file though (or keep it highly minimal), which I agree is becoming an anti-pattern.
The "open a different file" reasoning piece is a common pro-Tailwind statement and I do see the upsides.
I think that upside became more prevalent in the reusable components era, whereas previously CSS was targeting an entire HTML file (and thus the reasoning was more like SQL query than "this one element's styling").
With LLMs I think this upside is much smaller now though.
It seems that everyone is forgetting the web inspector as a tool for designing web pages. You can tweak properties and styles in a live environment, and then transfer your preferences to the css files.
Tailwind crazy adoption is something that makes me happy to nowadays be doing mainly boring stuff in distributed cloud systems and agents, instead of WebUIs.
For me Svelte and LLM completely removed my need for Tailwind. Turns out I was using it primarily to avoid CSS collision, and (to me) more logical syntax, rather than the self-imposed constraints.
Vue single-file components (SFC[0]) still works this way.
<template>
<!-- Largely just HTML -->
</template>
<script setup lang="ts">
// JS/TS as you would expect
</script>
<style scoped>
/* Component scoped styles here */
</style>
Very clean, easy to understand, and (as someone who started hand writing DHTML) it still feels very much like DHTML with more convenience and modern affordances.
Relying on React or Typscript in LLM era seems very stupid, just have the LLM setup whatever dom manipulation you want and have it write decent JS without slop. Far more offline compatible development almost negligible supply chain issues as well. At least ones you can control.
Layout design issues are orthogonal to choice of language and framework. You can apply the article's approach to plain static pages and to SPAs.
I tend to work closer to the latter end and find that both React and Typescript are extremely helpful to make my code extensible and maintainable. YMMV.
This works great for small sites/apps, but really starts to fall apart if/when it gains complexity where React starts to make sense. I've tried a few times to "just use plain javascript" with the LLM and initial results are often much better, but if the site grows a bit too complex, the LLM starts making a lot of mistakes and it can be hard to reason about as a human and get it on the right track. That hasn't been the case with the React apps IME.
I agree, specially for simple apps. it's much easier to upgrade if you are not relying on 3rd party or NPM's. Don't have to worry about code injections.
What a strange take. LLMs produce plausibly correct output, which is exactly where plain JavaScript and DOM manipulation will result in a spaghetti mess.
Frameworks like React that add structure to the data flow, component encapsulation, and a huge repertoire of patterns to train on, plus Typescript for immediate compile-time feedback loops… those are what LLMs thrive on.
> I got curious about what writing more semantic HTML would feel like.
I've been teaching semantic HTML / accessible markup for a long time, and have worked extensively on sites and apps designed for screen readers.
The biggest problem with Tailwind is that it inverts the order that you should be thinking about HTML and CSS.
HTML is marking up the meaning of the document. You should start there. Then style with CSS. If you need extra elements for styling at that point, you might use a div or span (but you should ask yourself if there's something better first).
Tailwind instead pushes the dev into a CSS-first approach. You think about the Tailwind classes you want, and then throw yet-another-div into the DOM just to have an element to hang your classes on.
Tailwind makes you worse as a web developer from a skill standpoint, since part of your skill should be to produce future-proof readable HTML and CSS that it usable by all users and generally matches the HTML and CSS specs. But devs haven't cared about that for years, so it makes sense that Tailwind got so popular. It solved the "I'm building React components" approach to HTML and CSS authoring and codified div soup as a desirable outcome.
Tailwind clearly never cared about any of this. The opening example on Tailwind's website is nothing but divs and spans. It's proven to be a terrible education for new developers, and has contributed to the div soup that LLMs will output unless nudged and begged to do otherwise.
A few counterpoints:
Treating markup and styles separately is great, in principle, but you'll always need additional markup for certain things. We knew this going back to the early 2000s.
There is nothing about Tailwind itself that forces you to use divs and spans instead of the appropriate HTML tag.
Documents and interfaces are different. Tailwind makes a lot more sense for interfaces. You can use Tailwind for the interface and scoped HTML selectors for other content.
Tailwind is around 4x faster and has practically no overhead compared to writing a complex CSS codebase. Whatever you think of it, this is always a benefit in its corner.
While I agree I do think there's some "aspiration of purity/correctness" in your approach that I've long let go of.
I look at the royal mess that is HTML/CSS/JS as a necessary evil, required when we want to target browsers. To me it's "just the presentation layer".
In my work I put a lot more emphasis on correctness in the db schema, or business logic in the backend.
When it comes to the messy presentation layer I prefer to write a little as possible, while still ending up with somewhat maintainable code. And for this Tailwind fits the bill really well: LLMs write it very well, new devs understand it quick, and it's quite easy to read-back/adjust the code later.
I 100% agree a Tailwind project is not the best way for a new dev to learn HTML/CSS. But then I prefer the new dev to focus on great db schemas, intuitive APIs, test-able biz logic, etc. Fiddling with the mess that's HTML/CSS is not the place where I consider human attention is best spent on (or where developers pick up skills to become much better developers).
This isn't about "purity/correctness" it's about the real experience of a blind person. Accessibility means caring about the HTML.
Your comment only mentions developers as the audience of HTML authoring, as opposed to users, which is a common attitude and the core problem with Tailwind.
I use Tailwind and have all kinds of "screen reader" directives in my templates.
Not sure if it helps, but if we get our first blind user I will gladly make some admends to make it more usable for them.
It seems that Tailwind is now blamed for the mess that is HTML/CSS. Tailwind certainly allows for accessible designs; it may not be the ideal solution, sure, but what we aim for is "good enough".
What's a good source to learn how to develop like this - to create HTML / CSS structure that's accessible?
EDIT: ignore. I can see you have some links in your profile. Will check it out.
> Tailwind instead pushes the dev into a CSS-first approach. You think about the Tailwind classes you want, and then throw yet-another-div into the DOM just to have an element to hang your classes on.
I wholeheartedly disagree. That mindset is not caused by Tailwind, but by being ignorant.
You can perfectly create an HTML document with semantic meaning and the add Tailwind just as any other CSS framework or pure CSS to it.
And DIVs do not carry meaning, they are specifically to add functionality or styling, so you can throw in as many as you like. Using them abundantly isn't good style, but the way you make it sound that they're evil isn't good either.
You're not wrong, and I mostly agree with you. I die inside when I see the div soup that a lot of sites have become. However, I think there is value in being able to have the important parts of CSS merged into the HTML a bit. Where that line is, is certainly up for debate (and I don't have the answer), but I've found a lot of my tailwind sites are more readable to me than my pre-tailwind sites, often because I don't have to context-switch and open a different file to be able to reason about the styling on an element. For big stuff the second file can be nice, but there's a lot of style tweaking that is great to be able to do right there in the HTML. Tailwind does really lead you to ignore the css file though (or keep it highly minimal), which I agree is becoming an anti-pattern.
The "open a different file" reasoning piece is a common pro-Tailwind statement and I do see the upsides.
I think that upside became more prevalent in the reusable components era, whereas previously CSS was targeting an entire HTML file (and thus the reasoning was more like SQL query than "this one element's styling").
With LLMs I think this upside is much smaller now though.
It seems that everyone is forgetting the web inspector as a tool for designing web pages. You can tweak properties and styles in a live environment, and then transfer your preferences to the css files.
So, we're just going a full circle back now. Interesting.
Tailwind crazy adoption is something that makes me happy to nowadays be doing mainly boring stuff in distributed cloud systems and agents, instead of WebUIs.
For me Svelte and LLM completely removed my need for Tailwind. Turns out I was using it primarily to avoid CSS collision, and (to me) more logical syntax, rather than the self-imposed constraints.
I'm lucky to have learned the web with Angular 2.x
It scopes CSS to components by default, and keeps HTML, CSS and JavaScript seperate.
Vue single-file components (SFC[0]) still works this way.
Very clean, easy to understand, and (as someone who started hand writing DHTML) it still feels very much like DHTML with more convenience and modern affordances.[0] Vue SFC docs: https://vuejs.org/guide/scaling-up/sfc.html
I am so happy that the only time I have to touch css anymore is for simple internal tools and pico is usually enough for them.
Relying on React or Typscript in LLM era seems very stupid, just have the LLM setup whatever dom manipulation you want and have it write decent JS without slop. Far more offline compatible development almost negligible supply chain issues as well. At least ones you can control.
Layout design issues are orthogonal to choice of language and framework. You can apply the article's approach to plain static pages and to SPAs.
I tend to work closer to the latter end and find that both React and Typescript are extremely helpful to make my code extensible and maintainable. YMMV.
This makes no sense. LLMs and agents benefit from (good) abstraction as much as humans do.
This works great for small sites/apps, but really starts to fall apart if/when it gains complexity where React starts to make sense. I've tried a few times to "just use plain javascript" with the LLM and initial results are often much better, but if the site grows a bit too complex, the LLM starts making a lot of mistakes and it can be hard to reason about as a human and get it on the right track. That hasn't been the case with the React apps IME.
I agree, specially for simple apps. it's much easier to upgrade if you are not relying on 3rd party or NPM's. Don't have to worry about code injections.
I have these two https://reddit.premii.com and https://hn.premii.com/ both works without any changes. Reddit will stop working once they kill the apis but until then it will work.
What a strange take. LLMs produce plausibly correct output, which is exactly where plain JavaScript and DOM manipulation will result in a spaghetti mess.
Frameworks like React that add structure to the data flow, component encapsulation, and a huge repertoire of patterns to train on, plus Typescript for immediate compile-time feedback loops… those are what LLMs thrive on.