Thoughts and feelings around Claude Design
AI-driven design tools like Anthropic’s new Claude Design are reigniting debate over how product UIs should be created and who should own that process. Many compare it to Figma, criticizing Figma’s heavy, proprietary workflow, rising prices, and weak integration with code, while praising Claude Design’s ability to generate working interfaces directly from prompts—but also warning about homogenized “vibe-coded” apps and current usage limits. Beneath the tool preferences is a broader shift: front-end, design, UX, and product roles are blurring as AI makes basic web app production cheaper and faster, raising questions about long‑term demand for traditional designers and developers.
Figma’s Pain Points and Business Choices
- Many commenters vent about Figma’s complexity (variables, components, props), performance issues on large files, and awkward handling of complex UIs like data grids.
- Pricing and seat-based licensing are heavily criticized as anti-collaboration and increasingly hostile to casual or educational use.
- Its proprietary, non-open file/protocol format is seen as a strategic mistake in an “agentic” era where tools that expose markup/code are easier for AI to use.
- Some argue Figma prioritized becoming an enterprise SaaS platform over being a great design tool.
Claude Design: Promise and Early Impressions
- Seen as a strong technical demo: good at quickly generating multiple UI variants, restructuring layouts, and handing off code-ready assets.
- Tight integration with Claude Code is praised; moving from mockup to implementation can be very fast.
- Usage limits are viewed as very restrictive; described as a “plaything” or research preview rather than a production tool.
- Not good for logos or illustration; focused on product UI and CSS/SVG generation.
Design–Code Gap and “Source of Truth”
- Many recount the long-standing Photoshop/Sketch/Figma → CSS/Storybook → app pipeline as lossy, duplicative, and ambiguous.
- There’s strong desire for tools where the design canvas is directly tied to real markup/code, reducing handoff and interpretation.
- Debate over whether Figma (or any design tool) can really be the “source of truth” versus the running app/code.
Shifting Roles and AI’s Impact
- Several argue front-end, UX, design, and product are converging, with AI enabling fewer people to cover more ground.
- Some report not writing much frontend code for months, or entire teams dramatically increasing output using AI assistants.
- Others counter that LLM-generated apps often have poor architecture, messy CSS, performance/maintainability issues, and require expert oversight.
AI Design Quality, Homogenization, and Limits
- Concern that “vibe-coded” UIs are simple because the underlying products are simple; AI may struggle with airplane-level design complexity.
- Worries about homogenous, same-y interfaces, though some welcome more predictable, consistent UIs.
- Many note that core UX problems (information architecture, edge cases, accessibility, platform conventions) remain hard and are not solved by prompting alone.