That's great. Started using Base UI early on via 9ui [0] and found the primitives very pleasant to work with, especially if one wants to compose more complex components from other Base UI components. Maybe Shad can reduce some of the dependencies they rely on now.
I think the two complement each other perfectly and will continue to do so. I keep writing AGENTS.md files for soft rules and custom linter rules for hard ones which IMO is the best of both worlds.
How about leveraging llms to produce deterministic codemods? You can then iterate on this by running the codemods and using other deterministic guardrails, feeding the results back into the llm to improve the codemods?
Yeah exactly this is right. In the same way that we don't get an LLM to write the HTML for each request, but we do get an LLM to write the code to make the HTML
I have used and mostly like Shadcn, and yet their Radix-based radio button was a bit much, as are other choices, where similarly overblown solutions were used.
For boring applications - do people prefer the copy paste approach of shadcn instead of a traditional ui library like mantine?
The copy paste approach may be easily modifiable but creates new problems - ie now there is an upgrade ai agent for something that should just be ticking up a version number.
Vendoring your components gives you the best of both worlds. You get a full component library but retain the ability to modify them as you want.
Your AI agent claim doesn't make any sense either. When upgrading normally your component just gets rewritten on disk. When switching from radix to base ui, a more comprehensive approach is needed.
I love Mantine. It’s easy to use, full featured with lots of components and helpers, and yet if you need to, you can use it headless and customise it too (I’ve never bothered though; I did add my own additional spacing/sizing options via the theme support though)
Any direct comparisons to base UI? I've been pretty satisfied with base UI so far, but my usage has been very basic so I'm open to switching if Mantine offers something better.
I’m leaning towards vendoring for all my new projects.
Grabbing an off-the-shelf UI library is easy in the short term, but it’s usually overcomplicated, implements things I won’t ever need, is hard to tweak if/when you want to distinguish your app from the thousand others using the same library, and when you do decide to upgrade it, all your tweaks break in subtle ways.
What I think would be the best approach is building your own UI library. You own it, you get to reuse it across different projects and maintain the same visual style (if desired), and you add features when you need them.
It's one more thing to maintain, and it's also difficult to push back on things. If you use off the shelf components it's much easier to say to designers and managers that a UX pattern is not available or not valid. You can point to the mature well known community owned UI library you use and make it authoritative. It's harder to do it if you build your own, suddenly each designer and developer is throwing things in there, adding features etc. It's also difficult to agree on the structure, so the components are well thought out, flexible, but also not so flexible to lose semantics. It's not an easy job, do you use slots, composition, rendering callbacks, there are too many decisions and you spend time building the UI library instead of actually shipping features.
Odd to me when I read stuff like this but also posts about how AI is making everything trivial. Surely a thing that almost every company did in the early 2000s should not be hard today, but of course today you can’t just write HTML + CSS, you must consider every front end framework introduced in the last decade.
I remember the 2000s web, it was primitive. It's not hard to combine some html and css, it is hard to support more complex dynamic actions and components, all with consistent theming and behaviors and have them composable and reusable.
It is true, yeah. But it allows you to own the brand identity, which is kinda useful if you have a brand.
I think we’ve found the holy grail with one of my clients recently. Our UI kit follows https://mui.com/ wherever makes sense, but we implement the components ourselves. This means (1) we don’t have to make too many architecture decisions – we just do whatever MUI does, and (2) it’s fairly easy to push back against adding features that don’t add a lot of value and deviate too much from, well, whatever MUI does.
To me, that's the value of shadcn (conceptually) - it's nothing more than a tool to bootstrap your own UI library.
It let's you copy in snapshots and change/edit/refine as you go. You own the UI kit the moment you copy it in. I see Shadcn 'upgrades' as nothing more than a stranger's PR to my UI kit, which usually means useless noise, unless there's a specific feature/bug that I need.
Honestly that’s about the only thing I like in shadcn. It makes scaffolding your UI kit extremely easy, but then you own it and can extend it in whatever way makes sense for you. Unfortunately I’m allergic to Tailwind, and React-only makes it a no-go for me, too :(
There are alternatives, of course, but what I’d like to see is a kinda unified component API spec that you can implement however you like, which both humans and AI can pick up without having to learn whatever idiosyncratic props you might have chosen. So I guess, I’d like to see other libraries use shadcn props with “sane”¹ implementation under the hood?
(¹ – in my case, just plain old Svelte components with inline CSS and/or CSS modules :-)
If you don't need to make any changes, it should be very simple to just upgrade by replacing the components. And if you need to make changes then well it's not gonna work with a traditional UI library.
> that should just be ticking up a version number.
Ah, but it's rarely just that in many systems. It can only be just that if the component library does exactly what you want. Unfortunately, it happens quite often that component doesn't entirely do what's needed.
People bolt on extra CSS to the components all the time. Two lines of CSS is very tempting if the alternatives are a few hours of work at least. But those two lines need to be verified against every new feature of the component library.
Do those two line fixes a lot, and upgrading becomes A Project.
> now there is an upgrade ai agent for something that should just be ticking up a version number.
If a component as basic as a button or a list view ever requires an “upgrade”, something is fundamentally wrong to begin with. HTML5, ARIA, etc. aren’t cutting edge technologies that the ecosystem still needs time to figure out. This should be pull once and forget.
Ticking up a version number is all fine and good until it requires a dependency upgrade you aren’t ready for. If for example you wanted to upgrade MUI from 4 to 5, you’d find react 17 wasn’t supported. And if you weren’t ready or able to upgrade react, then you’d just be stuck using a UI library going more out of date by the day.
With shadcn / the copy paste format, you’ll almost never see that happen. The button shadcn provides for example is just css / tailwind. And if you did ever for some reason want to bring in a dependency for your button component you wouldn’t have to consider it’s affect on your other UI elements. The rest of your components can live independently (for the most part)
We have customized UI components we got from shadcn and now some use radix and some use base ui, and some have other dependencies or no dependencies at all. Properly tree shaken this is not a big deal at all and we can upgrade components individually as needed.
For boring applications this may be a bit much. But even then if you wait too long and mantine falls behind more than a couple versions, who knows how easy it would be to get your whole project up to date.
Oh yes you are right. My bad. I got my dependencies backwards. We were using react 17 and wanted to upgrade to 18, but first we had to migrate an entire app on MUI v4 to v5 first since v4 was not supported by 18. And it was an upgrade full of breaking changes.
My point still stands that having a UI library like MUI deeply ingrained into your app can cause dependency and other issues like this. The speed you get upfront can bite you if you don’t plan carefully. Sure there are steps we could have taken to avoid such a mess. But that’s not how things often work out.
Still not sure if the alternative of vendoring like Shadcn, where a migration like this requires a black box probability machine to do the migration for you is any better.
My understanding is they made an LLM skill for updating because it’s trendy and people are gonna point an agent at it anyway. I think the winning idea is you can update components in your design system piecemeal instead of needing to do a Migration.
I never understood the copy paste thing. Shadcn just reeks of a fad, and even after doing copy paste and trying it, I didn’t see what the big deal was. But enjoy it, y’all. One of the benefits of AI is that HN isn’t clogged with talk of JavaScript frameworks as much
I think it depends on the project, but I've found that using installable libraries becomes a bottleneck over time as you find yourself running into their limits, or trying to customise them in ways they really aren't design for.
Personally I'd prefer React Aria; I don't have the patience to try understand why suddenly everyone is rushing to adopt Base UI instead of Radix which itself was once suddenly the thing everyone rushed to instead of Stitches.
Having a library not in anyway related to that to me feels like a big pro.
I have found React Aria to be very good. I really like how its a set of hooks, a set of premade components using said hooks, and I like how you can choose bits of either approach for your own components. Some of the hooks are very useful.
Both attempts[1] to surface this on HN failed but if you are using a PrimeTek product you need to be aware of this change.
PrimeNG, PrimeReact, and PrimeVue are all impacted and ongoing licencing will be $800 per developer. It's not the pro add-ons either this covers everything in the core libraries too. [2]
PrimeTek is no longer the maintainer of the project PrimeFaces with it now developed and maintained by independent volunteer developers who are not employees of PrimeTek.
I'm building Lily Design System as a response to Shadcn and similar systems not yet being available directly for multiple stacks including Svelte, Angular, Nunchucks. It's all free open source. If you want to send me your feedback, suggestions, etc. I can easily add things to Lily.
I'm in the same boat. We've been using PrimeNG for several apps at my job, and until we find a suitable replacement, we'll be stuck on using Angular 21.
That being said, a group did fork PrimeNG, and plan on maintaining it (new name pending)
It uses Zag js for its primitives which is framework agonistic as opposed to Radix or Base UI. Shadcn is only for React. If you use Solid.js for example, Skeleton is an option for your design system.
I hate the marketing-selling-linkedn style as much as anyone, but I don't think it's an LLM thing in particular. It's a style that existed before LLMs and it's very easy to make LLMs avoid it with one or two prompt paragraphs.
For what it's worth, I didn't get that vibe reading this post.
> Just read the post and it didn’t sound AI to me.
Those short and punchy two-part sentence groups very much feel like the writing that Claude does, like: The writing feels familiar. Suspicion earns its keep. Ultimately, the judgement remains yours. Not conjecture, your thoughts.
Then again, I bet how much aversion people feel to that sort of thing depends on how much they’ve been exposed to that, especially in frustrating circumstances. Personally, that’s a lot (daily Claude Code) and sometimes that writing makes me really upset.
People do write like that. Claude learned it from somewhere, after all. I read an X post yesterday where someone was complaining about "genuinely" as being an AI tell.
I think the broader phenomenon with the AI tells is it is revealing about a person's consumption. If they already interacted with and read material that resembled AI output, it wouldn't seem as weird. But if you encounter a particular pattern with the AI before you encounter the human patterns that trained it that way, it seems like an AI quirk.
> But if you encounter a particular pattern with the AI before you encounter the human patterns that trained it that way, it seems like an AI quirk.
It might also be the overuse of specific phrases and patterns. I sometimes scan over my own blog posts to see what appears too often and there are things that I overuse in my own writing as well.
It's not that it's bad writing. It's that people are used to bad writing. People are used to garbage. For example, your comment is awful from a literature or writing perspective. It needs to be edited. This comment that I'm writing right now is awful. We're used to crap writing and people who don't care about editing and putting out good, thoughtful words. When all you're used to is garbage, things that actually are better are generally going to seem odd and out of place. However, for those people that are used to better, more elegant writing, it's going to seem fine.
Of course, the difference here is context. In a comment, you're not expecting well-written sentences, structure, and editing. So we jump at these things that seem out of place because of the context.
> sometimes that writing makes me really upset.
You know what makes me upset?
- No writing
- Or bad documentation
- Or just no documentation
- Or just nothing being written down about something
For me, at its core, the most important thing is accuracy. Is it accurate? If so, good. We can start from there. If your issue is style, fine, but that's a personal judgment. As long as it's accurate, I'm fine.
That’s a genuinely smart take, and to be honest with you – most people never achieve this level of awareness. /s
Most humans speak in “garbage” and not perfectly correct sentences. Imperfection makes it human.
In this case, it’s not about unwanted sophistication, but fluff and specific sentence structures that tell you this was written by AI, not the actual people behind the project. I guess the closest analogy is to always being left on voicemail.
Claude/ChatGPT is not good or elegant writing. Read a book or a short essay from a respected author and it becomes obvious why different styles of writing are actually discernible. AI-generated text feels void of life and unintentional, it is legitimately a chore to read.
I call it "being shot in the face with bullet points". I was trying to use an LLM to write up some guidelines on how to use a piece of tech, and I kept getting frustrated that it felt like a slide deck rather than sincere persuasion.
Maybe it's a style that's always existed in moderation, but now it feels like it's being applied to every paragraph in every document or social posting.
Oh, I know exactly what you mean. I use plannotator with claude a lot and have much better time, since I asked for a specific styleguide.
I used "CD era MSDN reference and Raymond Chen blogging style" as a starting prompt for the styleguide and my work ability to digest AI plans raised a lot.
Couldn't recommend it more. Humble, insightful and respecting the reader
Exactly! See this is what I don't get about the Claude/GPT-style writing that's so prevalent everywhere and why it annoys me even more. It's just so easy to get rid of it, that's why it feels even more disrespectful when I still see it everywhere in full force, just a few sentences is enough to get rid of so much of the extremely obvious tells, sometimes even a "Don't write like an AI" in the prompt leads to completely acceptable results. Everywhere I use LLMs I put a tiny style-guide with instructions such as yours, and the results are just so much more pleasant I barely understand why it seems that almost everyone else seems incapable of it?
The obvious response is of course, they're just completely unbothered by it. Why change it if it doesn't even matter (to me)? I presume the set of people who use AI like this for writing and the set of people who are annoyed by it are largely not overlapping, and there is a possibility that a lot of the text I read and think sounds human, might be written by an LLM with a style-guide like mine. Still though, if 5 words genuinely can reduce annoyance by a lot of people who read your article, why does it feel like so many people haven't picked up on it yet? Or is the LLM writing highly loved & popular amongst other people perhaps?
I'm asking to scan projects on gitlab, go through some docs to find more grounding material, write a subarticle (in the same style), scan logs on the test env, issue some curls, etc.; until the whole article is digestible - in the "backing knowledge graph" department.
They are, and I am. While I don't use the words "LLM slop", I do have the urge to instantly stop reading any piece of writing that was obviously default Claude output with no effort to make it sound even remotely human written.
I'd rather read natural sounding, non-repetitive, and actually useful LLM text than the majority of reddit comments (including the serious ones) for instance.
I think you are forgetting about the decency and dignity part of "respect".
Yes there's a quality component to the role of communication in how it respects other people.
There's also honesty, transparency, truth and vectors along the dimensions of "Are we claiming and presenting the truth or are we bending facts and creating impersonations and warping reality?" Most AI is used for the latter today: people are having AI's write their words and speech for them and then the AI says things as though it were the human like "I said xyz" when the AI is NOT the human who did those things. That's lying and deception and disrespect to the reader.
I may complain that I don't like the way a sleezy con man talks, and I may be able to detect his communication patterns, but that doesn't mean I want the con man to speak in a different way I can't detect as sleezy. I don't want to talk to the con man.
Obfuscating LLM output to trick the reader into thinking it wasn't LLM output is not respectful.
Disclaimer, I only use it to grow the "knowledge hub".
It's a single git project at my $USER home, that is referenced in global memory. It contains as much information about work things, as possible, to be productive.
I found that, if I allowed Claude to create the notes, it actually became more and more useful, but without the guideline, I just could barely get through reading it manually.
Isn't that a bit like working in the movie industry and no longer being able to enjoy films because you are always thinking about how they were made and noticing their flaws?
More like working in the movie industry and noticing all the places where the director took the cheap shortcut instead of putting effort in. And while the result isn't much worse, it still feels a bit disrespectful. You reminisce about the time five years ago when things were much more human
Oh, and the target audience of movie you are watching are other people in the movie industry. Nearly everyone in the audience can tell where production cheaped out
The post effectively communicated what it needed to. It seemed both written and structured in a way that was optimized for human consumption.
If they used AI to write this and that gave them more time to volunteer their time towards developing this fantastic piece of open source software, then this all seems like a good thing to me.
It is clearly not effectively communicated if the tool used to write it stamps its signature in such a way that it distracts from the great work they’re doing.
All writing is open to interpretation by the reader.
Some people let punctuation or grammar mistakes ruin an entire text or post for them, because they choose to focus on those flaws, rather than looking past them and taking in the content itself.
"this was written/assisted by AI" is starting to feel like next-gen "this has spelling/grammar mistakes in it, therefore it is invalid".
First time I’ve seen this take, but man it rings true.
I remember having teachers in middle school who seemed draconic to me in their emphasis on proper grammar and punctuation, however as an adult I’m lucky to read an email that contains paragraphs, much less proper punctuation.
Times are changing, and it’s up to the individual to decide how they want to respond to it
This doesn't even account for the fact that while the tech industry's dominant language is English, lots of people in our industry count English as a second (or third/fourth/etc) language, so they might use AI to help them write in English in an effort to be better understood by others.
To arbitrarily dismiss all AI-assisted writing shows a real lack of understanding of the possible reasons one might use it beyond the "lazy author" trope that everyone likes to throw out there.
I think the problem is that too many people use AI to write bullshit. Thus AI writing becomes a cue for bullshit.
For people who have internalized AI writing as a cue for bullshit, it is very difficult to read obvious AI writing without constantly being cue-ed that the thing they're reading is bullshit. Even if it's not.
The issue is that you can't trust that every word of the message was intentional due to how much Claude will usually add various levels of nonsense, some more obvious some less to your voice or requests. Considering how much Claude alters the meaning of what I want to communicate I wouldn't trust the accuracy of that output right.
And I use Claude a lot, 24/7, but not for things like that. And I appreciate how much it elevates my productivity, but not like this. It usually prioritizes or highlights the wrong things, it overfixates on one thing I said and adds random content there out of nowhere.
So then I can't tell what part of it is slop, how slop, and it becomes impossible to trust.
LLMs literally are incapable of writing in the way humans are. Generating a string of tokens is a completely different task than composing prose be it fiction, documentation, or product releases. It would take an insane harness on the par of a complexity of a coding harness to write at the level of college graduate.
So when I see someone saying LLMs are suitable for this use I must assume that they don’t think what they are writing is worth the effort or they don’t understand how effective written communication works. Either way, I’m resentful of it in the way someone reviewing a slop PR is.
I’ve been thinking about this a lot this week after having fable try to write some design docs and it outputting 1k line docs which I had to manually rewrite to 200-300 lines. I should of just did it myself to start.
Then don't trust it. Don't read it. Don't use it. But I don't get this obsession with trying to police and control everyone's use of AI. Hint: you can't control everyone. They're going to do what they want whether you like it or not. The only thing you can do is move on.
Am I the only one who expects ALL writing to be generated or at least edited by an LLM going forward? It's like pointing out the fabric in my clothes was not hand woven by a human. Sounding like an AI wrote it is not as valid a criticism as "it was unclear about...", or "It was too long", or "It left out this important point...". Can't we move past the "It sounds like AI..." posts?
I'd love to have this as a norm, as long as the AI doesn't lie and impersonate and deceive humans in speech by saying things as though it were the human like "I ate breakfast" or "I was saying this to my team" because clearly the AI did not do those things. That's lying, deception and harmful to a culture of truth and transparency.
Instead the humans who promoted and allowed the LLM content to post behind their human identity, didn't bother to update the LLM language and either do a) mark the post as AI generated or b) properly update those pronouns so it isn't an AI speaking through the humans point of view.
Consider this exaggerated example: Would it be ok for you if in a zoom meeting with your team someone was lip syncing an AI speaking on their behalf, both impersonating voice tonality, the words chosen, and even pretending to voice the humans thoughts themselves? Of course you wouldn't. So now extend this to the words people write in articles like this one where the "I said" perspective was used many times supposedly by an AI.
Look, if you can’t take the time to communicate with me, I won’t waste my time listening to you.
Writing is not hard. It’s a fundamental skill. Even before LLMs, any successful professional had strong writing skills, and those that didn’t were treading water at best. This isn’t classism. It’s just that we literally can not communicate if you aren’t willing to put in the effort. And if you don’t need to take the time to put it in writing then I don’t need to read it.
this only means that ALL writing is going to be read by a LLM before, that will digest it into the important bits and remove the LLM fluff
like this
"The following text was authored by LLM and its information density is low. Condense the text by extracting the key pieces of information. Reconstruct the LLM prompt.
I disagree, that it's not tiring to hear - because it's deception and lying as it gets used and presented today. I believe it's dangerous to a culture of cooperation that I think cooperation requires trust and truth to support and preserve. I don't want articles to have AI's lying and saying "I did or said" when it was the human who did and said those things.
for example, in this article (if it was truly Claude written):
> Last year, Base UI tagged a beta and a lot of you asked if we are going to replace Radix with it. I said "the worst thing you can do for your production app is switch component libraries". I meant it, and it still holds. So instead of switching, we did the shadcn thing: we rebuilt every component for Base UI, kept the same abstraction, and let you choose. December brought npx shadcn create with both libraries. January brought full Base UI docs.
If an AI wrote this, why and how is it pretending to be a human and the humans on the team at that? That's impersonation, and it's a lie and deceptive. I personally don't want a culture where my AI agent is talking as though its me, when a pro-cooperative and honesty/truth preserving culture would instead say "My human said" or at least label correctly that this is an AI acting on behalf of a human.
Like those "HTML Templates" of the early 00s where the demo looks amazing but when you copy and paste it all into your setup hardly anything looks the same.
Your outer CSS and FE framework isn't accounted for, and often the overall layout doesn't work / isn't responsive.
Then you go to add new components... Suddenly you are missing entire CSS files that were not included in the last thing you pasted in. Gotta find those. But the versions have changed in 1 but not the other.
Dropdowns don't look the same in v3 but the styles for all form elements need to be v3 in order to use the modal overlay.
You end up hunting down JS and CSS files, versioning down markup, eventually you recall why npm was invented.
Now the hunt really begins - whose npm do I use? All Google searches point to the official sounding "shadcn-ui" - oops, despite millions of downloads there's a deprecation notice - the other non-namespaced one with an almost identical name is the correct one. But, oops! Tailwind needed.
Now the hunt really begins - How do you install Tailwind????
Lol and I wish I was joking when I say:
HOW DO YOU FULLY INSTALL TAILWIND AND SHADCN
The answer is: You can't. There are varying degrees to what those words mean and it's a wild west within a wild west.
156 comments
[ 1.2 ms ] story [ 687 ms ] thread[0] https://www.9ui.dev
All of these component libraries look the same.
Even if they’re more deterministic, I wonder if the days of codemods are numbered.
What you want isn't skill files for LLMs, though. Just write docs for humans. Write a migration guide, for humans.
It's going to take us a while to realize they should be the same things, skill files and docs.
https://news.ycombinator.com/item?id=46688971
I’m trying out Ark UI on a side project. They do have some genuinely useful components, like tags input: https://ark-ui.com/docs/components/tags-input
They have a tabs/“segment group” component with a nice animated active element indicator which would probably be tricky to implement: https://ark-ui.com/docs/components/segment-group
And then they also have stuff like overcomplicated “click to copy” button and a <details> reimplementation: https://ark-ui.com/docs/components/clipboard, https://ark-ui.com/docs/components/collapsible
All with a verbose markup that renders as a div soup.
The copy paste approach may be easily modifiable but creates new problems - ie now there is an upgrade ai agent for something that should just be ticking up a version number.
Vendoring your components gives you the best of both worlds. You get a full component library but retain the ability to modify them as you want.
Your AI agent claim doesn't make any sense either. When upgrading normally your component just gets rewritten on disk. When switching from radix to base ui, a more comprehensive approach is needed.
When You're Ready to Migrate
You don't need to migrate. But if you want to, we built a skill for it:
pnpm dlx skills add shadcn/ui
Then ask your coding agent:
migrate accordion to base-ui
Thanks internet.
Grabbing an off-the-shelf UI library is easy in the short term, but it’s usually overcomplicated, implements things I won’t ever need, is hard to tweak if/when you want to distinguish your app from the thousand others using the same library, and when you do decide to upgrade it, all your tweaks break in subtle ways.
What I think would be the best approach is building your own UI library. You own it, you get to reuse it across different projects and maintain the same visual style (if desired), and you add features when you need them.
It's one more thing to maintain, and it's also difficult to push back on things. If you use off the shelf components it's much easier to say to designers and managers that a UX pattern is not available or not valid. You can point to the mature well known community owned UI library you use and make it authoritative. It's harder to do it if you build your own, suddenly each designer and developer is throwing things in there, adding features etc. It's also difficult to agree on the structure, so the components are well thought out, flexible, but also not so flexible to lose semantics. It's not an easy job, do you use slots, composition, rendering callbacks, there are too many decisions and you spend time building the UI library instead of actually shipping features.
I think we’ve found the holy grail with one of my clients recently. Our UI kit follows https://mui.com/ wherever makes sense, but we implement the components ourselves. This means (1) we don’t have to make too many architecture decisions – we just do whatever MUI does, and (2) it’s fairly easy to push back against adding features that don’t add a lot of value and deviate too much from, well, whatever MUI does.
It let's you copy in snapshots and change/edit/refine as you go. You own the UI kit the moment you copy it in. I see Shadcn 'upgrades' as nothing more than a stranger's PR to my UI kit, which usually means useless noise, unless there's a specific feature/bug that I need.
There are alternatives, of course, but what I’d like to see is a kinda unified component API spec that you can implement however you like, which both humans and AI can pick up without having to learn whatever idiosyncratic props you might have chosen. So I guess, I’d like to see other libraries use shadcn props with “sane”¹ implementation under the hood?
(¹ – in my case, just plain old Svelte components with inline CSS and/or CSS modules :-)
Ah, but it's rarely just that in many systems. It can only be just that if the component library does exactly what you want. Unfortunately, it happens quite often that component doesn't entirely do what's needed.
People bolt on extra CSS to the components all the time. Two lines of CSS is very tempting if the alternatives are a few hours of work at least. But those two lines need to be verified against every new feature of the component library.
Do those two line fixes a lot, and upgrading becomes A Project.
If a component as basic as a button or a list view ever requires an “upgrade”, something is fundamentally wrong to begin with. HTML5, ARIA, etc. aren’t cutting edge technologies that the ecosystem still needs time to figure out. This should be pull once and forget.
With shadcn / the copy paste format, you’ll almost never see that happen. The button shadcn provides for example is just css / tailwind. And if you did ever for some reason want to bring in a dependency for your button component you wouldn’t have to consider it’s affect on your other UI elements. The rest of your components can live independently (for the most part)
We have customized UI components we got from shadcn and now some use radix and some use base ui, and some have other dependencies or no dependencies at all. Properly tree shaken this is not a big deal at all and we can upgrade components individually as needed.
For boring applications this may be a bit much. But even then if you wait too long and mantine falls behind more than a couple versions, who knows how easy it would be to get your whole project up to date.
My point still stands that having a UI library like MUI deeply ingrained into your app can cause dependency and other issues like this. The speed you get upfront can bite you if you don’t plan carefully. Sure there are steps we could have taken to avoid such a mess. But that’s not how things often work out.
Is it ever that simple?
I use Material UI for all my AI coded applications. Am I doing something wrong? What difference would Shadcnd even make in a world of Claude Code?
Isn’t either infinitely customizable via CSS? So what’s the difference?
Not a boring application, a very large application -- each major version update was a tedious process because they completely upended many APIs.
We've migrated to shadcn, and upgrades are now easy; we can upgrade a single component -- add the new ones free.
With MaterialUI we had to update EVERYTHING to their new APIs to be able to take advantage of the new features anywhere.
With shadcn we can be selective.
There were a couple of rough migrations though.
I wrote something about it a few years ago when shadcn was relatively new on the scene https://andrewingram.net/posts/recipe-kits-a-great-alternati...
Mantine sucks so much it's unreal.
It's the react-table of UI libraries. Absolutely over-engineered and terrible in every way compared to everything else available.
Trying to decide between the two atm.
Having a library not in anyway related to that to me feels like a big pro.
I have found React Aria to be very good. I really like how its a set of hooks, a set of premade components using said hooks, and I like how you can choose bits of either approach for your own components. Some of the hooks are very useful.
It even looks incredible when building desktop apps. We used it to build DB Pro [1] and the DB Pro website, and everyone compliments us on our design.
I see it becoming the defacto choice for UIs especially when building with agents.
[1] https://dbpro.app
PrimeNG had a licensing change recently and I'm looking at a suitable alternatives for a fresh project.
PrimeNG, PrimeReact, and PrimeVue are all impacted and ongoing licencing will be $800 per developer. It's not the pro add-ons either this covers everything in the core libraries too. [2]
PrimeTek is no longer the maintainer of the project PrimeFaces with it now developed and maintained by independent volunteer developers who are not employees of PrimeTek.
[1]: https://hn.algolia.com/?q=The+Next+Chapter+of+PrimeTek
[2]: https://primeui.dev/nextchapter
Any community forks?
https://lilydesignsystem.github.io
That being said, a group did fork PrimeNG, and plan on maintaining it (new name pending)
https://github.com/openng-org/open-prime
https://basecoatui.com/
For what it's worth, I didn't get that vibe reading this post.
Would be kind of funny if someone from the team came out and said it was written by a human.
Those short and punchy two-part sentence groups very much feel like the writing that Claude does, like: The writing feels familiar. Suspicion earns its keep. Ultimately, the judgement remains yours. Not conjecture, your thoughts.
Then again, I bet how much aversion people feel to that sort of thing depends on how much they’ve been exposed to that, especially in frustrating circumstances. Personally, that’s a lot (daily Claude Code) and sometimes that writing makes me really upset.
I think the broader phenomenon with the AI tells is it is revealing about a person's consumption. If they already interacted with and read material that resembled AI output, it wouldn't seem as weird. But if you encounter a particular pattern with the AI before you encounter the human patterns that trained it that way, it seems like an AI quirk.
It might also be the overuse of specific phrases and patterns. I sometimes scan over my own blog posts to see what appears too often and there are things that I overuse in my own writing as well.
Of course, the difference here is context. In a comment, you're not expecting well-written sentences, structure, and editing. So we jump at these things that seem out of place because of the context.
> sometimes that writing makes me really upset.
You know what makes me upset?
- No writing
- Or bad documentation
- Or just no documentation
- Or just nothing being written down about something
For me, at its core, the most important thing is accuracy. Is it accurate? If so, good. We can start from there. If your issue is style, fine, but that's a personal judgment. As long as it's accurate, I'm fine.
Most humans speak in “garbage” and not perfectly correct sentences. Imperfection makes it human.
In this case, it’s not about unwanted sophistication, but fluff and specific sentence structures that tell you this was written by AI, not the actual people behind the project. I guess the closest analogy is to always being left on voicemail.
Maybe it's a style that's always existed in moderation, but now it feels like it's being applied to every paragraph in every document or social posting.
I used "CD era MSDN reference and Raymond Chen blogging style" as a starting prompt for the styleguide and my work ability to digest AI plans raised a lot.
Couldn't recommend it more. Humble, insightful and respecting the reader
The obvious response is of course, they're just completely unbothered by it. Why change it if it doesn't even matter (to me)? I presume the set of people who use AI like this for writing and the set of people who are annoyed by it are largely not overlapping, and there is a possibility that a lot of the text I read and think sounds human, might be written by an LLM with a style-guide like mine. Still though, if 5 words genuinely can reduce annoyance by a lot of people who read your article, why does it feel like so many people haven't picked up on it yet? Or is the LLM writing highly loved & popular amongst other people perhaps?
When people say LLM slop is disrespecting the reader, I don't think they are complaining about style.
I'm asking to scan projects on gitlab, go through some docs to find more grounding material, write a subarticle (in the same style), scan logs on the test env, issue some curls, etc.; until the whole article is digestible - in the "backing knowledge graph" department.
- Understanding what you wrote
- Verifying correctness of any claims
- Putting in at least as much effort writing as your audience will reading
- Not sharing generated content if you can't do the above. If you must, then explicitly disclaiming your use of AI
It does not mean "prompting AI better"
I'd rather read natural sounding, non-repetitive, and actually useful LLM text than the majority of reddit comments (including the serious ones) for instance.
slop just means "I don't like this style"
when AI writes more reliably in a way that people do like, they will stop calling everything AI does slop.
Yes there's a quality component to the role of communication in how it respects other people.
There's also honesty, transparency, truth and vectors along the dimensions of "Are we claiming and presenting the truth or are we bending facts and creating impersonations and warping reality?" Most AI is used for the latter today: people are having AI's write their words and speech for them and then the AI says things as though it were the human like "I said xyz" when the AI is NOT the human who did those things. That's lying and deception and disrespect to the reader.
Obfuscating LLM output to trick the reader into thinking it wasn't LLM output is not respectful.
It's a single git project at my $USER home, that is referenced in global memory. It contains as much information about work things, as possible, to be productive.
I found that, if I allowed Claude to create the notes, it actually became more and more useful, but without the guideline, I just could barely get through reading it manually.
I'd never publish anything with such origin.
Oh, and the target audience of movie you are watching are other people in the movie industry. Nearly everyone in the audience can tell where production cheaped out
If they used AI to write this and that gave them more time to volunteer their time towards developing this fantastic piece of open source software, then this all seems like a good thing to me.
Some people let punctuation or grammar mistakes ruin an entire text or post for them, because they choose to focus on those flaws, rather than looking past them and taking in the content itself.
"this was written/assisted by AI" is starting to feel like next-gen "this has spelling/grammar mistakes in it, therefore it is invalid".
I remember having teachers in middle school who seemed draconic to me in their emphasis on proper grammar and punctuation, however as an adult I’m lucky to read an email that contains paragraphs, much less proper punctuation.
Times are changing, and it’s up to the individual to decide how they want to respond to it
To arbitrarily dismiss all AI-assisted writing shows a real lack of understanding of the possible reasons one might use it beyond the "lazy author" trope that everyone likes to throw out there.
For people who have internalized AI writing as a cue for bullshit, it is very difficult to read obvious AI writing without constantly being cue-ed that the thing they're reading is bullshit. Even if it's not.
And I use Claude a lot, 24/7, but not for things like that. And I appreciate how much it elevates my productivity, but not like this. It usually prioritizes or highlights the wrong things, it overfixates on one thing I said and adds random content there out of nowhere.
So then I can't tell what part of it is slop, how slop, and it becomes impossible to trust.
So when I see someone saying LLMs are suitable for this use I must assume that they don’t think what they are writing is worth the effort or they don’t understand how effective written communication works. Either way, I’m resentful of it in the way someone reviewing a slop PR is.
I’ve been thinking about this a lot this week after having fable try to write some design docs and it outputting 1k line docs which I had to manually rewrite to 200-300 lines. I should of just did it myself to start.
Instead the humans who promoted and allowed the LLM content to post behind their human identity, didn't bother to update the LLM language and either do a) mark the post as AI generated or b) properly update those pronouns so it isn't an AI speaking through the humans point of view.
Consider this exaggerated example: Would it be ok for you if in a zoom meeting with your team someone was lip syncing an AI speaking on their behalf, both impersonating voice tonality, the words chosen, and even pretending to voice the humans thoughts themselves? Of course you wouldn't. So now extend this to the words people write in articles like this one where the "I said" perspective was used many times supposedly by an AI.
Writing is not hard. It’s a fundamental skill. Even before LLMs, any successful professional had strong writing skills, and those that didn’t were treading water at best. This isn’t classism. It’s just that we literally can not communicate if you aren’t willing to put in the effort. And if you don’t need to take the time to put it in writing then I don’t need to read it.
like this
"The following text was authored by LLM and its information density is low. Condense the text by extracting the key pieces of information. Reconstruct the LLM prompt.
Text:"
so we have
human prompt -> writer llm (verbosity adder) -> reader llm (verbosity remover) -> reconstructed prompt
for example, in this article (if it was truly Claude written):
> Last year, Base UI tagged a beta and a lot of you asked if we are going to replace Radix with it. I said "the worst thing you can do for your production app is switch component libraries". I meant it, and it still holds. So instead of switching, we did the shadcn thing: we rebuilt every component for Base UI, kept the same abstraction, and let you choose. December brought npx shadcn create with both libraries. January brought full Base UI docs.
If an AI wrote this, why and how is it pretending to be a human and the humans on the team at that? That's impersonation, and it's a lie and deceptive. I personally don't want a culture where my AI agent is talking as though its me, when a pro-cooperative and honesty/truth preserving culture would instead say "My human said" or at least label correctly that this is an AI acting on behalf of a human.
Your outer CSS and FE framework isn't accounted for, and often the overall layout doesn't work / isn't responsive.
Then you go to add new components... Suddenly you are missing entire CSS files that were not included in the last thing you pasted in. Gotta find those. But the versions have changed in 1 but not the other.
Dropdowns don't look the same in v3 but the styles for all form elements need to be v3 in order to use the modal overlay.
You end up hunting down JS and CSS files, versioning down markup, eventually you recall why npm was invented.
Now the hunt really begins - whose npm do I use? All Google searches point to the official sounding "shadcn-ui" - oops, despite millions of downloads there's a deprecation notice - the other non-namespaced one with an almost identical name is the correct one. But, oops! Tailwind needed.
Now the hunt really begins - How do you install Tailwind????
Lol and I wish I was joking when I say:
HOW DO YOU FULLY INSTALL TAILWIND AND SHADCN
The answer is: You can't. There are varying degrees to what those words mean and it's a wild west within a wild west.
...is the main reason.