The CSS Zen Garden dream, finally shipped(josprague.com) |
The CSS Zen Garden dream, finally shipped(josprague.com) |
The Bright Creative site is also AWOL.
Too bad neither is set up to redirect to the current https://daveshea.com/
Hope he's doing well. Guys like him, Doug Bowman, Jeffrey Zeldman... all a huge influence on teenaged-me hacking away on my little blog and thinking of going to SXSW someday.
EDIT: He's just left after being at Shopify for six years: https://lnkd.in/p/dcYBvQfq
nails on a chalkboard. when will agents stop writing like this
For me it was here. I have tried for months to get AI to write for me and have now fully embraced the fact that LLMs are not ready to do that for me. None of them. I’ve gone back to writing my own content and only using the LLM to brainstorm.
The point is - now the modern vanilla web platform really can support first-class CSS-styled app components, from design systems, in production. This is really cool!
HTML and CSS were originally designed as document-oriented languages for digital document display; think MS Word or PDF with cross-document hyperlinks, and not an app SDK. The idea was simply "style your web site's titles, headings, paragraphs, footers, etc. in just one place; change it there and the whole site will update". That's the dream that CSS Zen Garden was meant to showcase. Interactivity and web apps came later, and CSS has taken a very long time to catch up.
But catch up it has, and I'll be glad when frontend web development sheds Tailwind and React in favor of the updated vanilla web platform; and I won't complain if we use super light web component libraries like Lit[0]).
[0] https://lit.dev
Human readability is a very useful proxy for good modularity and separation of concerns (and modularity increases scalability and reusability). Since LLMs take the whole thing in at once, they are terrible at module boundaries. (They also suck at theory of mind, and their comments reflect it, but I digress.)
Search for "flare" if you're looking for the CSS.
If your team is rethinking its front-end foundation, get in touch.
Instead of thinking of separation of concerns as a religious principle, think about what benefits it actually yields in the case of HTML.
When the markup is completely separated from its styling, this means that the CSS rules need a way to be mapped onto the DOM. This involves specific elements, organized a specific way. If anything is out of place in the markup, the properties may not apply as intended. Similarly, if the CSS doesn't account for DOM structures, then you need modify rules or add new ones.
This way of mapping is an implicit structure. It's something extra that's required, beyond the source code that's there. It can be computed by the browser, but it's not explicitly stated for a developer looking at the code to understand.
Contrast this with Tailwind, Tachyons, etc., where nearly all the structure is explicit, less complex, discoverable, and out in the open.
Hidden structure looks nice to the developer. The code "looks" clean. But understanding a code base's implicit structure is ephemeral, and there's inherent technical debt that will need to be repaid later.
Tailwind looks ugly, but there are no hidden abstractions or structures in the code, beyond its easy conventions and a minimal config file.
And, yes, there are ways to make high-abstraction CSS work, say, with a component library of known DOM patterns. But at that point, why not build the styles into the components directly, a la Tailwind, instead of adding complexity?
Perhaps the article isn't about font import, those better be in html headers or simply stick to system fonts.
For applications people spend a expected to spend a lot of time in, the user may even be able to select the theme.
But, things like CSS custom properties have eaten into some of the basic use cases like palette swaps, assuming they're supported in the default theme.
I'm so tired of AI writing
So very quickly web designers started to just hard-code pixel widths into the CSS, removing any dynamic sizing.
For example, on this very page: https://csszengarden.com/ the two-column layout is done by using two overlapping DIVs with a padding on the left one. You could do 3 columns by using the same trick with left padding.
Want to do four columns? Tough. You'd better redesign your whole HTML for that to work. Or just hardcode the sizes.
Of course, this quickly led to interfaces where buttons have text that is cut off. But with lots and lots of useless whitespace padding around them.
Happy to answer questions about the layered CSS, the legacy browser fallback, how the tokens got exported from design files, or anything else you're curious about.
Always weird to see what posts really bring out the negativity on HN, but pay that no mind. Some people can't stand others being happy (and some are just having a bad day...)
Be proud of what you accomplished and it must feel great to have contributed to something like Firefox!
1. Go to site
2. Visually scan the page for the style controls
3. Click one
4. Everything changes
5. Visually scan the page for the style controls etc.
I dont want to visually scan every time I change the stlye.
Just give me a view and let me operate on things. We're going to have our "LCARS" moment soon. Where no one really needs to create a UI - the system creates one for you at time of use - based on what you're looking at and doing. We're no longer going to need UI/UX, everything will adapt to the person using things.
It reminds me of people wanting Hollywood UIs that would obviously suck to actually use. Just completely ignoring the reality of using what you are suggesting.
People still aren't understanding that the mid-game of AI is the end of fixed cultural objects.
Everything becomes potentially customisable by the user, including the affordances, the UI, and the content algorithms.
The dream of the CSS Zen Garden never really extended beyond itself. That wasn't the point. The point was to show people "look at all these pages that look nothing like each other: they're all the same HTML." That was the magic.
Keep in mind the context: you still largely needed structural HTML to effect the layout of your page. Almost everybody was building with HTML where the markup was critical to the appearance. Frontpage, Dreamweaver, and every other tool generated almost no CSS—or at least no good CSS.
This isn't my interpretation. From the site itself:
> There is a continuing need to show the power of CSS. The Zen Garden aims to excite, inspire, and encourage participation. ... CSS allows complete and total control over the style of a hypertext document. The only way this can be illustrated in a way that gets people excited is by demonstrating what it can truly be, once the reins are placed in the hands of those able to create beauty from structure.
It was pretty obvious even back then that they were just people who didn’t want to learn something new and they would not listen to anything pro-CSS at all. In fact, developers who insist LLMs are not useful for development remind me of them a lot.
Anyway, the CSS Zen Garden was released, and it seemed to shut everybody like that up overnight. It conclusively won the argument that CSS sites could look as good as table-based sites and could even do things the table-based sites couldn’t.
Verily, Tailwind is the assembly language of CSS. No structure, no semantics, no abstractions, only twiddling specific bits of visual representation.
To turn to it is to admit that your product lacks the structure and the design vision that allow to use some abstraction ("semantic classes"), and that all you can usually do is to patch some areas of it, disconnected from the rest, and unable to touch the rest (lest it goes down in flames). Assembly is definitely suitable for patching.
This is often the endgame of old large products that changed hands and directions many times, without much care.
Maybe there's an overwhelmingly positive case to be made for high-abstraction CSS with hidden structures. Outside those few select cases, we've known the tradeoffs for 25 years, and the added complexity is rarely worth it.
It has. It's called @apply. It works fine. The Tailwind author is wrong about recommending people to avoid @apply. The Tailwind community at large is mostly wrong about it too (except people that like and use @apply).
I don't know how anti-abstraction came to become popular, but if you like semantic classes you can still use Tailwind. (why use Tailwind? It has some nice defaults and a coherent design language)
https://css-tricks.com/tailwinds-apply-feature-is-better-tha...
The anti-abstraction crowd, that push for repeating the same boilerplate over and over again, honestly grinds my gears. Even assembly doesn't make writing boilerplate its ideology (it has macros, and actual subroutines)
Most developers and organizations don’t want to and they don’t want to think why or what doing so would buy them. They will probably find the tailwind local maxima enabling that reasonably comfortable and may even repeat the familiar mantra as has been done here. It especially pairs well with unreflective product management, which is the most popular kind of product management.
CSS Zen Garden was a proof of concept built around a single markup file. But if you extend that concept to hundreds of dynamic pages and a well built CSS design system, that's where the real power of CSS shows.
> In real life, it doesn't really work.
In real life it works incredibly well, which is what CSS Zen Garden originally showed, and what we've proven by applying these concepts to Firefox.
> If anything is out of place in the markup, the properties may not apply as intended. Similarly, if the CSS doesn't account for DOM structures, then you need modify rules or add new ones.
When stylesheets are written with an understanding of the Cascade and specificity, this isn't the case at all.
People have this idea that they should be decoupled. You can do that if you really contort them. Naturally, they are coupled.
The structure of HTML puts constraints on the style. You have to know about the style when writing HTML, even in this Zen Garden project. You can't escape the style and HTML understanding each other. If you have a compelling reason to make a broad range of styles swappable then OK - but you are constraining the design space considerably and there is usually no reason for this complexity and reduction in expressiveness.
The document and the style depend on eachother. You see this with localization. You might need to change spacing or padding or breakpoints to accomodate different languages (text being different length in different languages).
This is why all UI paradigms tightly couple style and structure. QT, GTK, SwiftUI, Flutter, etc. Its the same with CSS/HTML except for people who delude themselves into thinking these were supposed to be loosely coupled.
I’d previously understood the relationship to be that the document could remain unchanged, whilst the style (including visual structure/layout) changed. The simplest canonical example is that you can change the font, and all the words and meaning remain unchanged.
> You see this with localization. You might need to change spacing or padding or breakpoints to accomodate different languages (text being different length in different languages).
This is a really intereting case. Strictly I’d say it was a different document, if the language has changed. Really good translations change idiom, which can change meaning, and words, at the level of ohraes, although the intent would be preserved.
But I’m nit-picking, being contrarian, or exploring the edges of your comment.
Today, many web documents are app GUIs. Each field is essentially the same “thing” regardless the languge of its label. And in that sense, these are the same document, and your point holds.
I still think that if you can keep the markup as minimal as possible (reduce the div nesting to just what’s require to mark up logical sections of the content), then you could make a good stab at Zen-gardening your design. It’s very hard to do, however.
And how all six people links are to LinkedIn, when most of them have websites. (Wouldn’t surprise me if some of the LinkedIn profiles were incorrect, but since you can no longer view anything on LinkedIn without having an account…)
It's okay to exercise nuanced views (as well as choosing to say nothing.)
what if it was the difference between not getting the thoughts out there and being able to celebrate the fruits of their labor?
The "difference" is that of spending five entire minutes proofreading the output. But this blog post, like the other 98%, is once more the tasteless fruit of very, very little labor. Whether you value it or not, whether you are even able to recognize it anymore, writing for an audience is a skill.
I (and probably GP) feel legitimate dismay when I read the output of someone who had an obvious opportunity to exercise good taste for the benefit of the people who actually take the time to read it.
The continued inconvenience to you of reading opinions like this is obviously extremely unfortunate.
> [..] , and on a site like this one that matters more than architectural purity. [..] Nothing in it depends on a build step [..] That distinction is worth keeping in mind [..] What matters is [..]
All that "inspiration" resulted in a loss of scalable interfaces and a switch to hard-coded element dimensions. And this persisted for a LONG time because Web UI designers are, in general, bad and were fine with flashy UI that is finger-breakingly bad.
Heck, it's _still_ a problem because people are by now completely oblivious to that. The art of automatic UI layout has been lost completely.
> One of the reasons we moved to XHTML, semantic markup, CSS instead of Tables... was to enable accessibility.
Again, nope. Try a design with 3-4 columns with the old CSS and see how it behaves. The order of elements in the HTML often differs from the visual order, confusing the readers.
It also breaks tab-navigation, even the https://csszengarden.com/ has the "View All Designs" button that is inaccessible.
Then add absolutely-positioned headers and other popular CSS patterns.
I enjoyed working with a semantically styled frontend project as recently as 2024. It also used a React component library, which helped insulate per-component styles, while sharing the common parts.
That’s exactly how it should be. If you serve me up AI-generated rubbish that I’m then supposed to sift through, you’ve earned the ridicule of the whole world.
Implies you'd accept it if it wasn't rubbish. Seriously doubt your filter is based on quality.
> you’ve earned the ridicule of the whole world
As if human writing immunizes you from that. Published an article about garbage collection and had people complain about my human voice. One went so far as to suggest I had mental issues. I received praise too but some things you just never forget.
Best part is Claude warned me repeteadly that I should tone it down, and I deliberarely ignored it in order to preserve my voice and keep it pure. I'm much more inclined to listen now.
The posts above claiming design and content are inextricably linked are probably blessed with all the standard senses.
As usual it's a bunch of folks talking past each other. Style and content together is a nice simplification when you want to conform to a particular vision. Separation is a nice abstraction if you want to enable multiple views of the same content (like a pdf version or a colorblind version or a vision impaired version) without having to change/control the content.
Like most things in this space, different tradeoffs make one better than the other in different scenarios. Taking a hard stance of "one true way" is probably missing some imagination or experience.
Accessibility certainly requires a content-first approach, I agree with you there. If you view each part of an application solely in terms of what it looks like to a sighted user, it's easy to miss the subtleties of what the underlying structure ought to be.
But when you start layering on the presentation, you're still generally binding that presentation layer tightly to the current content. If you, for example, were to restructure the links in the Zen Garden footer for some reason, most of the styles would break because they're bound to the current structure of the content. The separation you talk about only exists in one direction.
By comparison, when I talk about separation and decoupling in terms of code, I mean that there is a single, explicit, and tight API that acts as a clear boundary between two units. By explicit, I mean that I've designed the units specifically so that they expose this API - this is almost exactly the opposite of what I should be doing with content and presentation, as you point out! By tight, I mean that the API is as minimal as I can make it - two units interacting via an API need to understand nothing of each other except that API, and that API is as small as possible.
None of that is happening when it comes to the interaction between content and presentation. The content is not explicit - I'm actively trying to write the content for the design, which precludes making it an explicit API surface. Nor is the content tight - again, I want to build the content without worrying about how I will layer the design on top, and I don't want to limit myself to just a small surface area here.
So to me, it makes sense to see the content and presentation as being coupled together, even if that coupling ideally goes in one direction only. And if two systems are coupled together, it makes sense to do that explicitly by co-locating those systems. Therefore, I tend to prefer systems like Tailwind that let me apply the presentation layer directly in the content layer. (In practice, I mostly prefer other tools that still let me write real CSS syntax, but Tailwind is pretty damn convenient a lot of the time.)
As an aside, talking about accessibility, I'd forgotten how inaccessible some of the Zen Garden designs are with their addition of extra content in the presentation layer alone.
Consider: MathML has a presentational schema and a content schema. The whole point of presentational MathML is the intermingling of concerns. If you want interoperable data, you use content MathML.
Interoperability is a noble goal, but if you want HTML to do everything you're already barking up the wrong tree.
Yeah. And you cant effectively do that.
Even this project proves it. You have to know which CSS features never to use. You know, like CSS3, which started in 1998.
It's gonna be a no brainer for a lot of people.
But it’s also possible that they have other, more important stuff going on in their life? Or that they find writing legitimately difficult?
I’m not excusing anything, or saying that you can’t feel dismay or even sickened.
I was merely proposing another perspective, that’s all.
It’s just not a very efficient strategy, it might work for some readers, but as the comments show it’s a big turn off for many. Where it says “There’s one honest footnote”, that’s really grating, like the other commenter I also stopped reading there.
That’s not necessary rational, it’s just that we are social creatures that are really sensitive to those kind of surface level cues in communication, to gauge credibility. I’m sure many authors today use LLM in brainstorming, research, ask it to steelman an argument etc and no readers complain because it doesn’t impact the authorial voice.
The problem then is with letting LLM ‘fleshing out the details’: you end up having the LLM’s style and words all over your text. You can’t really do that without your authorial voice loosing credibility. I think next time, you have to go over it and edit it and make it sound like you.
If you don’t want to that, the good news is that I think people will start to get less critical of sloppy writing, as long as it sounds human. So you could sound convincing while leaving some details raw and not have the LLM have a go at it.
Now try to navigate something like the WSJ/NYTimes home page.
If you remember the old days of good UI, there was this idea of "tab groups".
To this day, this is STILL not working well for DIV-based designs.
Imagine that I write an article manually and keep it to myself. Then I prompt my favorite AI model and it somehow generates the exact same article, against all odds, despite having never seen it before. Then I publish the AI's article, which is equal to mine. Is the article rubbish?
> I can send a prompt myself, don't waste my time republished AI amswers.
I can do it too. I consider it rude to reply to people with AI answers during conversation. Not because I hate AI, but because it made the social interaction pointless.
That doesn't somehow invalidate the value of an AI generated article though. Not everyone thinks to prompt it with that exact idea. Not everyone has an AI they can prompt. Not everyone wants to waste their limited tokens on the idea.
I realize that reading Claudish can be aggravating but that doesn't somehow eliminate the ideas contained in the article. Those can still be very valuable, which is why I advocate for quality filters.
"Substantially" to me would mean not your ideas and not your writing. I'm sure there are borderline cases that are hard to decide but a rule of thumb for me would be if it takes more time to read than it took the "author" to write.
It's possible for the ideas to come from the author while the writing is outsourced to someone else. Ghostwriting is an established practice.
Effort involved in writing sounds a lot like the "sweat of of the brow" doctrine, which is rejected even by copyright monopolists.
I can sympathize with the notion that AI writing currently sucks, but not that it's automatically invalid or deserving of mockery. Maybe one day we'll have far better models. Maybe they'll do a good job without inserting any load-bearing vocabulary.
Asking for the prompts is fine, but might not be sufficient. The process iterative, and may incorporate data that won't be published. For example, I recently published a Sol/Fable benchmark article. It would be trivial for anyone to prompt a new version of the article if they had the raw benchmark data, but this data will never be published.