Where the stack is grinding.

• On 6 January 2026 Tailwind Labs laid off three of its four engineers, with revenue down close to 80% while the framework sat at roughly 75 million npm downloads a month.
• Adam Wathan, in a GitHub comment: Tailwind is growing faster than it ever has and is bigger than it ever has been, and our revenue is down close to 80%.
• The cause is not a competitor. AI assistants generate Tailwind without anyone visiting the docs, which is where the paid products were discovered.
• Wathan declined a proposal to make the docs more AI-friendly, on the grounds it would accelerate the same collapse.
• The argument that followed splits two ways: that venture-funded AI products monetise open source without paying maintainers, and that the sponsorship income made the layoffs avoidable.
• The part that generalises past Tailwind: any project whose funding depends on people reading its documentation now carries the same exposure, and the ones that do not are the ones worth betting a stack on.

• devclass · the layoffs
• Socket · OSS business models
• the-decoder · the structural warning

Checked 27 Aug 2026

• Tailwind v4 is itself built on the modern CSS it is accused of papering over. Its own release notes name native cascade layers, registered custom properties via @property, and color-mix() as the foundation.
• The default palette moved off RGB in v4: "We've upgraded the entire default color palette from rgb to oklch, taking advantage of the wider gamut."
• The upgrade removed deprecated utilities outright rather than deprecating them — overflow-ellipsis becomes text-ellipsis, so the old class emits nothing instead of erroring.
• The default shadow, radius and blur scales were renamed so every utility has a named value: shadow-sm became shadow-xs and shadow became shadow-sm, with the same shift on rounded, blur, drop-shadow and backdrop-blur. Existing markup keeps compiling, at the wrong visual weight.
• JavaScript config files still work, but are no longer detected automatically — they must be loaded explicitly with @config — and corePlugins, safelist and separator are not supported in v4.
• On adoption, State of CSS 2025: Tailwind drew 2,041 responses against Bootstrap's 1,194, from the 3,977 people who answered the frameworks question.

• Tailwind · v4 upgrade guide
• Tailwind · v4 release notes
• State of CSS 2025 · other tools

Checked 27 Aug 2026

• Issue #45759 asks the question directly and has no settled answer: in a world of React 19, RSC, shadcn and Tailwind, is Pigment CSS still the future of MUI?
• Pigment CSS was announced, never shipped stable, and went back into refactoring for v1.
• MUI v7 still ships on Emotion, and Base UI launched without a Pigment story.
• The long-standing complaints sit underneath: 300kb+ gzipped at full library, roughly 12kb of Emotion runtime, and slower hydration under SSR.
• In development, barrel imports from @mui/icons-material can be several times slower than path imports because of the sheer number of icons.

• mui/material-ui#45759
• MUI · introducing Pigment CSS
• LogRocket · Pigment CSS in practice

Checked 27 Aug 2026

• MUI X moved from licensing the first ten developers to requiring a licence per developer — fifty developers now means fifty licences.
• Existing customers were grandfathered and can keep renewing on the legacy plan.
• The stated reason is keeping the price accessible for small teams, which does not address the teams the change actually hits.
• Issue #10244 records the confusion it caused about whether the per-developer price was capped.

• mui/mui-x#10244
• MUI · 2024 price update
• MUI X · licensing

Checked 27 Aug 2026

• Radix development slowed noticeably after the WorkOS acquisition, which put its long-term support in question.
• Base UI shipped stable 1.0 on 11 December 2025 with 35 accessible components and a dedicated maintenance team.
• Projects created with shadcn/create were already picking Base over Radix roughly two to one before the switch was official.
• In July 2026 shadcn/ui made Base UI the default for new projects.
• The uncomfortable part is not which library is better: whoever copied components into their own repo now owns a migration nobody will do for them.

• InfoQ · Base UI v1.0
• shadcn/ui · Base UI as the default
• tldraw#7584 · evaluating the migration

Checked 27 Aug 2026

• The recurring shape is not a crash. Things stop working without saying anything, which is the expensive class of bug.
• MUI TextField under Controller produces the uncontrolled-to-controlled warning — a structural tension, since RHF is uncontrolled-first and MUI is controlled-first.
• Validation silently does nothing when handleSubmit does not wrap the handler, or when a field was never registered.
• Omitting the generic on useForm drops type safety across errors, watch and setValue, with no error to tell you.
• watch across many fields degrades performance visibly on large forms.

• RHF#4956 · uncontrolled to controlled
• RHF#3940 · MUI uncontrolled input
• Discussion #8117 · watch performance

Checked 27 Aug 2026

• react-hook-form was not built for React 19 actions or server-side form handling, and the discussion tracking it is still open.
• The majority position is hybrid: RHF for client validation, React 19 and Next for the server side.
• The bridge people actually use: feed useActionState's returned state into useForm as errors and default values.
• The stronger position argues for dropping the dependency, because action-returns-state removes a whole category of sync bugs.
• The live disagreement is server validation latency against the need for immediate client feedback.

• RHF discussion #11832 · React 19
• Oberlehner · RHF with useActionState
• Wisp · which one to use

Checked 27 Aug 2026

• You stop debugging a component and start debugging a JSON document, which needs familiarity with the schema rather than the UI.
• Schema validation happens at runtime, so errors arrive while running instead of while compiling.
• Genuinely custom rendering still lives in JavaScript — the schema configures, it does not implement.
• Under cascading rules the page stalls; distributed per-field state is what keeps linkage-heavy forms responsive.
• Schemas with hundreds of fields become large documents that weigh on initial load.

• SurveyJS · scalable schema-driven forms
• Formily · distributed field state
• Form.io · forms from JSON schema

Checked 27 Aug 2026

• One published position calls exposing className in React components an anti-pattern that produces brittle, unmaintainable code.
• The opposite need is real: a design system without an escape hatch fails the first case it did not anticipate.
• MUI opened an issue specifically to standardise sx across its libraries; StyleX opened the mirror-image discussion about supporting className.
• Under Tailwind the same question turns operational — whose class wins when the consumer passes one? That is what tailwind-merge, cva and tailwind-variants exist to answer.
• The underlying trade-off is flexibility through className, style and sx against a closed design language with controlled overrides.

• dev.to · className as an anti-pattern
• mui#29816 · standardising sx
• stylex#442 · supporting className
• tailwind-variants · comparison

Checked 27 Aug 2026

• Role checks scatter through the codebase, and string comparisons plus role lists stop scaling as features multiply.
• There is no clean way to express hierarchy — dashboard granted but dashboard:checklist withheld.
• There is no standard way to say deny this specific thing while allowing its parent.
• Auditing is hard because decisions live in thousands of imperative ifs, duplicated across layers.
• Non-developers cannot manage access without a code change.
• The consistent warning: frontend checks are ergonomics, never security — they never replace the backend.
• The direction of travel is permission-based rather than role-based; the UI does not need to know roles.

• LogRocket · access control models
• Oso · frontend authorization
• Permit.io · RBAC best practices

Checked 27 Aug 2026

• React's own documentation makes the case: Effects are "an escape hatch from the React paradigm", and "Removing unnecessary Effects will make your code easier to follow, faster to run, and less error-prone."
• On chaining, the docs are explicit — the component and its children "have to re-render between each set call in the chain", and as requirements move "the 'chain' you wrote doesn't fit". The verdict: "Such code is often rigid and fragile."
• An Effect that immediately updates state restarts the whole render-and-commit cycle from scratch.
• The two cases the docs single out as never needing an Effect: transforming data for rendering, and handling user events.
• The separate community complaint is over-abstracted custom hooks that own several responsibilities without declaring any, then get reused where they were never designed to go.

• React · You Might Not Need an Effect
• React · useEffect reference
• Codeminer42 · custom hook problems

Checked 27 Aug 2026

• This one is a practitioner pattern rather than a measured finding — treat it as a shape people recognise, not a statistic.
• The reported failure is not writing validation twice, it is that the two copies diverge and nobody notices until production.
• The client validates enough for fast feedback; the server revalidates, sometimes more strictly, sometimes differently.
• A requirement changes and only one side is updated, because nothing links them.
• The shared-schema answer is well understood; what it depends on is a boundary discipline that is easy to state and hard to hold.

• Leapcell · sharing Zod across a monorepo
• Tecktol · shared schema in practice

Checked 27 Aug 2026

• A practitioner pattern rather than a measured finding — the shape is consistent across write-ups, the frequency is nobody's published number.
• The example that recurs: a designer changes the primary colour, and the developer has to hunt that value across several repositories.
• Designers use the newest variables assuming they match the codebase, while the value underneath is stale.
• Tokens designed only from the design side produce long names and values that do not map cleanly onto CSS properties.
• The convergent fix is a naming convention agreed before the system is built, not retrofitted after.

• Into Design Systems · tokens are hard
• Figma · design tokens

Checked 27 Aug 2026

• Retool's pricing page lists Free, Team, Business and Enterprise. On deployment it says Retool "runs in the cloud or self-hosted in your own VPC" and that "Enterprise plans support both options" — it does not state that Team or Business include self-hosting.
• Listed rates at the time of checking: Team €9 per builder and €5 per internal user, Business €46 per builder and €14 per internal user, Enterprise on request.
• The engineering complaint underneath is older than any vendor: multiply one ordinary CRUD screen by ten or twenty entities and the simple back office is a real application.
• That is the whole reason admin-panel frameworks exist — not to make CRUD interesting, but to stop rebuilding auth, tables and forms per entity.

• Retool · pricing
• Refine · Retool alternatives

Checked 27 Aug 2026