Jane Street Designer Adopts Claude AI for Prototyping, Bypassing Figma

Edwin, a designer on the options desk at Jane Street, has transitioned his primary prototyping workflow from traditional visual design tools like Figma to the Claude AI model. This shift, which began after he joined the firm last summer, was driven by the necessity to rapidly learn and implement new technical stacks, specifically OCaml and the Bonsai framework. By generating functional code prototypes directly within the codebase, Edwin bypasses the creation of static mockups and specification documents, allowing for immediate testing of user-facing features.

The designer reports that his reliance on Figma has "fallen off a cliff" over the past two months, with AI now being utilised for larger-scale projects, including prototypes with 2,000-plus line diffs. Previous attempts with other large language models, such as Copilot, Cursor, and Gemini, were described as disappointing, but Claude has proven indispensable for his current role. This approach enables "free, unlimited iteration," allowing for rapid tweaks to user interface elements, copy, and functionality without the typical engineering design back-and-forth.

A specific example of this workflow is the development of an LLM prompting feature for JSQL, an internal SQL dialect used across various user-facing tools at the firm. The prototype allows Edwin to refine details such as the submit button, keyboard shortcuts, and confirmation messages in real time. This method accelerates development and empowers designers to build working proofs of concept, making it easier for stakeholders to evaluate ideas by simply using the functional tool rather than reviewing static proposals.

To manage this new process, Jane Street has implemented a specific protocol for reviewing these prototypes. They are treated as "living proposal docs" where the code is considered disposable, and reviewers provide feedback on design and user experience rather than just code quality. Eventually, reviewers take over the idea and implement it in a separate feature, referencing the prototype but owning the production code. This approach aims to maintain a collaborative design space, ensuring that the prototype is viewed as an iterative tool rather than a final product.

Despite the efficiency gains, the designer expresses concerns that this iterative, AI-assisted process may constrain creative thinking and prevent "big changes" to ideas. He draws parallels to past debates about designers learning to code, where critics argued that programming limits creative scope. However, Edwin notes that without AI support, contributing to the new technical stack would have felt out of reach, and the ability to work directly in the medium has restored a sense of freedom and fluidity to his design process.