I design with Claude more than Figma now
6 hours ago
- #AI in Design
- #Prototype Workflow
- #LLM Integration
- The author was previously skeptical of LLMs due to disappointing results in areas they were already proficient in, such as game tweaks and design tasks.
- After joining Jane Street, AI became indispensable for learning new skills like OCaml and Bonsai and unexpectedly transformed their design workflow.
- Instead of creating spec docs and Figma mockups, they now build prototype features directly in code using Claude, enabling rapid iteration and real user testing.
- This approach eliminates ancillary work like Figma components and documentation formatting, focusing effort on improving the actual artifact.
- The workflow evolved from handling small UX fixes to tackling larger projects, including prototypes with 2000+ line diffs and skipping Figma entirely for some apps.
- Using AI allows designers to independently create working proofs of concept, making it easier to evaluate ideas and demonstrate feasibility without relying on engineers.
- A downside is that reviewers receive fully baked features, potentially limiting collaborative input, so they frame prototypes as living proposals for iterative feedback.
- There's a concern that designing with Claude may constrain creativity to iterative outcomes, similar to historical debates about designers coding with tools like React.
- Despite challenges, the author finds empowerment in returning to hands-on creation, feeling freer to experiment and work directly in the medium.