The “AI will replace designers” debate finally has substantive data. Both Claude Design (Anthropic’s design-focused product) and Figma Make can now generate working product mockups from natural-language briefs. We gave them the same prompt and compared the output.
The headline: neither tool replaces a designer. But they replace different parts of the workflow, and which one fits your work depends on what you’re optimizing for.
The brief
We asked both tools to design a screen for a hypothetical product:
A mobile-first dashboard for a Vietnamese fintech app showing the user’s current balance, last 5 transactions, and a “Send Money” CTA. Match a clean, trustworthy financial app aesthetic. Include both light and dark modes.
Identical brief, no follow-ups, single shot.
What Claude Design got right
Information hierarchy. The balance card sat at the top, large and prominent. Transaction list used clear date grouping. The CTA had pill-shaped emphasis without screaming. Spacing felt deliberate.
System thinking. Claude returned not just a screen, but a full token export — color scales, typography ramp, spacing scale, component patterns. The output was usable as a starting point for an actual design system, not just one screen.
Trustworthy aesthetic. The fintech category has tight conventions (banks, brokerages, fintech all converge on similar patterns). Claude correctly read the trust signals and avoided trendy decorative elements.
Where Claude struggled
The Vietnamese-specific cultural nuance was missing. Currency formatting defaulted to $ notation rather than đ. Names in mock data were anglicized. Number grouping followed Western conventions (10,000.00) rather than Vietnamese (10.000,00). These felt like training-data biases.
Visual polish at the pixel level also fell short — alignment was approximate, font weights inconsistent in places.
What Figma Make got right
Pixel-level precision. Figma Make produced a file with proper auto-layout, real components (not flattened groups), and consistent spacing tokens applied. You could open the result and continue working in Figma immediately.
Plugin/system integration. The output respected Figma’s design system constraints. Variables were properly assigned. The light/dark variants used Figma’s modes correctly.
Realism in mockups. The transaction list felt lived-in — varied amounts, realistic merchant names, time stamps that made sense. Less polished aesthetically, but more honest.
Where Figma Make struggled
Decision-making. Where Claude made bold typographic choices, Figma Make hedged. Type sizes were “safe” (14/16/24) without character. The result was professional but generic — it could ship for any of 50 fintech apps.
System-level reasoning. Figma Make optimized for the screen, not the system. Tokens were assigned but the underlying logic wasn’t articulated. If you asked “why this spacing scale?”, you wouldn’t get a coherent answer.
Side-by-side comparison
| Dimension | Claude Design | Figma Make |
|---|---|---|
| Speed to first output | 25s | 40s |
| Output editability | Tokens + spec | Native Figma file |
| Aesthetic boldness | High | Medium |
| Cultural fit (VN context) | Weak | Neutral |
| System coherence | Strong | Medium |
| Pixel-level polish | Medium | Strong |
| Best for | Direction-setting | Implementation |
Which to reach for
The decision tree is simple if you frame it right.
Reach for Claude Design when: You’re at the start of a project, need direction, want a system not a screen, or are working in a category Claude has good training data for.
Reach for Figma Make when: You’re mid-project with established patterns, need pixel precision, or want output that drops into your existing Figma file without translation.
In practice, the workflow we settled on uses both: Claude for the system definition (tokens, type scale, component patterns), Figma Make for the screen-by-screen execution.
AI tools don’t replace designers. They redistribute where designers spend time. Less pixel pushing, more decision making. — Internal team retro, after this experiment
Conclusion
Neither AI tool is “better.” They optimize for different parts of the design process — Claude for direction, Figma Make for execution. The interesting workflow shift in 2026 isn’t about replacing designers; it’s about designers becoming more directorial, treating AI tools as collaborators that handle the boilerplate so humans can focus on judgment calls.
The winner of “Claude vs Figma Make” is the designer who knows when to use each.