A Jane Street designer ditches Figma for Claude-built prototypes

A Jane Street designer describes how Claude has displaced Figma as their primary design tool. After years of skepticism toward LLMs—which underperformed on tasks they were already good at—they found AI indispensable when ramping up on unfamiliar territory like OCaml and Bonsai. The breakthrough was realizing that working prototypes in the actual codebase beat mockups and spec docs for nearly every kind of design proposal, from small UX fixes to 2000+ line diffs touching data models and libraries.

The workflow now looks like: write a brief problem description, hand it to Claude as a prompt, iterate against a running build, deploy to a dev environment for user feedback, and submit a PR-equivalent that already behaves the way they want. A recent prototype adding LLM prompting to an internal SQL dialect input went through dozens of refinements—keyboard shortcuts, button tweaks, copy adjustments—that would have stalled in engineering backlogs at a prior job. The author argues this levels a long-standing asymmetry: engineers can prototype their own ideas, while designers have historically had to lobby someone else to build proofs of concept.

The shift creates new friction around review. Handing engineers a fully built feature risks reducing them to code reviewers rather than design collaborators, so the team now frames these prototypes as disposable living proposals—the reviewer eventually reimplements the idea as production code. The author also worries that iterating with Claude locks them into a refinement mindset and away from bigger creative leaps, echoing the old ‘should designers code’ debate, but concludes that working in the real medium again feels more freeing than years spent in Figma.