AI Regeneration Is Killing the Business Case for Shared UI Libraries

A 2026 Tilburg University meta-analysis found a small but statistically significant homogenization effect in human-AI co-creation, robust across sensitivity analyses and not explained by publication bias. That finding anchors a new argument from Griffiths Waite's Daniel Curtis: when coding agents can generate most standard UI on demand, the company-wide component library — a decade of near-unquestioned engineering orthodoxy — buys far less than it once did.

Curtis, UI Development Manager at the Birmingham-based consultancy, writes that his team has spent most of the last decade maintaining React component libraries across enterprise clients and is now rethinking them. His position: keep the design system, tokens, guidelines and tests central, and let the shipped component code go.

What did the library actually provide?

A shared library bundled four promises: reuse, consistency, encoded expertise, and a single source of truth. Only reuse and encoded expertise justified shipped code — precisely what AI can now supply. Consistency and source-of-truth already belong to the design system and its design tokens, the shared color, type and spacing values a component merely references.

The real cost was never the initial build. "Every line of code you own, is a line you have to maintain, test, fix, and update," as a daisyUI piece on abstraction ownership puts it. A widely adopted Hacker News commenter, on his company's second library rollout, called internal libraries "expensive to produce and maintain" — "much more complex than your average CRUD app" — requiring low staff rotation and years-ahead thinking.

Why did regeneration get cheaper than reuse?

Marco Kotrotsos, in an essay titled The Death of the Component Library, describes how Tailwind's own paid-component business was undercut by its ubiquity: because Tailwind dominates training data, "when developers need a Tailwind component, they don't visit the documentation. They prompt. The AI generates the code directly." A 2026 technical report on component-based development for AI coding describes tooling that lets agents "inspect, modify, test, debug, and regenerate code fragments." Denis Uraev frames the shift as a move from reusable to regeneratable code, with tests that let agents "verify correctness independently."

Griffiths Waite already uses the Figma MCP server in projects and reports it "remarkably accurate" at generating UI directly from designs, combined with custom skills that keep generated code aligned to their standards.

What replaces the package?

A specification stack: design tokens pushed through Style Dictionary; machine-readable guidelines and accessibility rules versioned beside the code; and engineering principles encoded as searchable libraries rather than tribal knowledge. Verification runs through visual-regression tools — Chromatic, Percy, Playwright — with screenshot comparison against the Figma source and computed-style checks confirming that color, spacing and type resolve to tokens.

Versioning moves up the stack. Component rules, tokens and per-component instructions ship as versioned bundles that pin the Figma designs; teams regenerate affected components and run the test suite when versions bump.

Does AI sameness change the calculus?

The Northeast Times reported that two startup founders arrived at separate meetings with decks built in Anthropic's Claude Design tool that a designer said appeared "generated by the same company" — matching four-rectangle layouts, centered text. A separate 2024 ideation study found people "produce less semantically distinct ideas with ChatGPT."

Curtis counters with McKinsey's Business Value of Design study, which tracked 300 listed companies from December 2012 to December 2017: top-quartile McKinsey Design Index performers outpaced industry counterparts by 32 percentage points on revenue growth and 56 on total shareholder returns. An opinionated system — real type scale, spacing rhythm, motion — is the lever that pulls a model off its statistical average.

Where does a shared library still win?

Accessibility. Telerik's argument holds: "AI code generating tools do not excel at accessibility," and the fix is architectural — "use libraries that encode accessibility into their API contracts." A 2026 paper, Software Reuse in the Generative AI Era, warns that blindly trusting generated code is "conceptually not all that different from cargo cult development."

Curtis's recommendation splits shared code three ways: regenerate the high-volume, low-risk surface; curate the genuinely hard widgets — data grids, comboboxes, date pickers — as audited code; and keep the design system, tokens, guidelines and test suites central. For teams without a funded platform group, he argues, the shrinking of the shared library is not a loss worth mourning — and the design system, in that world, matters more, not less.