Skip to content
← All work Anthropic • Claude Design 2026

Teaching a model what good design looks like

Design prompt expert for Claude Design — prompt frameworks, evaluation of visual output, and UI/UX feedback on Claude’s web and mobile interfaces.

RoleDesign Prompt Expert (contract)
TimelineMar 2026 – present
ClientAnthropic
DeliverablesPrompt frameworks, evaluations, UI feedback
Overview

A model can generate a thousand layouts. Someone has to say which ones are good.

I work with product, research and engineering to refine the design output of Claude Design: writing and testing prompts, judging visual quality and coherence, and documenting frameworks so the work is repeatable across internal teams.

Prompt frameworks

Structured prompts that guide layout, hierarchy and brand consistency.

Evaluation

Scoring outputs for quality, coherence and alignment with design principles.

UI feedback

Accessibility and clarity notes on Claude’s web and mobile interfaces.

Process

Write, generate, critique, rewrite.

Each prompt goes through rounds of generation and critique. The interesting part is not the first output — it’s finding the phrasing that makes the tenth output as good as the best one.

Video · autoplay
Placeholder loop — replace with real examples where allowed.
Outcome

Systematised what used to be taste.

Documentation and frameworks now used by internal teams; measurable improvements in consistency of generated visuals.

Next projectTeaching a model what good taste looks likeOpenAI • Design evaluation 2026