Shadcn/UI now defaults to Base UI instead of Radix

Shift from Radix to Base UI

• Several commenters are positive on Base UI’s low-level primitives, finding them pleasant for composing complex components and hoping this reduces dependencies in shadcn/ui.
• A key claimed advantage: Base UI is more “headless” and less opinionated than Radix, which suits shadcn’s pattern (wrapping primitives in custom components).
• Some report friction adopting Base UI, especially around forms and reuse of existing patterns.

Copy‑Paste Model vs Traditional UI Libraries

• Many see shadcn’s copy‑paste vendoring model as a way to “own” the code, avoid brittle dependency chains, and selectively upgrade components without large breaking changes.
• Others find this model overhyped or outright bad: upgrades become manual work, risk divergence, and require new tooling (including AI “migration agents”) rather than just bumping a version.
• There’s concern that shadcn makes many apps look visually identical unless teams invest heavily in customization.

Comparisons Among UI Libraries

• Mantine is heavily debated: some find it “brilliant” (full-featured, customizable, with unstyled primitives); others describe it as over-engineered and painful in large projects.
• Base UI vs React Aria: some favor React Aria’s hooks-based, library-agnostic feel and dislike the churn from Radix → Base UI trends; others are drawn to Base UI’s newness and flexibility.
• Alternative ecosystems mentioned: Ark UI, MUI/Base UI, DaisyUI, Material UI, Skeleton (with Zag), Basecoat, and various “shadcn-like” ports for non-React frameworks.

Codemods, Migrations, and LLM Agents

• The project’s shift from codemods to LLM-based migration agents is seen as notable.
• Some argue LLMs and codemods are complementary: use LLMs to generate deterministic codemods plus lint rules.
• Others prefer human-readable migration guides and worry about over-optimizing for agents instead of developers.

AI‑Generated Writing in the Announcement

• Many commenters believe the changelog post has a characteristic LLM “voice”: punchy, LinkedIn-style, verbose, and formulaic.
• Some find this disrespectful or untrustworthy, arguing that AI prose often buries sparse information in fluff and may misrepresent who is “speaking.”
• Others are indifferent or supportive: they value skimmable structure, see AI-assisted writing as inevitable, and prefer humans focus on code rather than prose.
• There is disagreement over whether the main issue is style, accuracy, honesty about AI use, or readers’ own expectations.

Angular Ecosystem and PrimeNG Fallout

• For Angular users seeking a “shadcn-like” solution, multiple options are mentioned, with one in particular praised as clearly ahead.
• PrimeNG/PrimeReact/PrimeVue’s move to closed source and paid licensing draws concern; a community fork aims to continue an open alternative.

Semantic HTML and Complexity of UI Toolkits

• Some criticize modern UI kits (Base UI, Ark UI, etc.) for “div soup” and re-implementing things like <details> or copy-to-clipboard with heavy markup.
• This is seen both as overengineering and as a symptom of the web platform lacking richer native UI primitives.

General Sentiment

• Strong mix of enthusiasm and fatigue: people appreciate shadcn as a bootstrap for custom design systems but are wary of ongoing frontend churn, complex abstractions, and style homogenization.