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.
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.
Mantine is brilliant, I can build anything in it quickly and then extend it or completely customize the theme or individual components, but there is a learning curve. I would not call it a giant learning curve.
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 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.
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?
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.
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.
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
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.
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.
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
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.
Even if they’re more deterministic, I wonder if the days of codemods are numbered.
[0] https://www.9ui.dev
Trying to decide between the two atm.