I design with Claude more than Figma now — Jane Street designer workflow

Edwin Morris, a designer at Jane Street, describes how Claude Code has largely replaced Figma in his design workflow. Instead of spec docs, Figma mockups, and proposal reviews, he now builds prototype features directly in the codebase using Claude Code. The workflow: write a problem description and proposal → open editor with Claude → prompt with the description → get a working prototype → iterate → push to dev environment → get user feedback → submit a feature (PR).
Key quote: “A prototype feature in the actual codebase has felt better in almost every way compared to mockups and docs.” Morris reports that Claude Code gives “free, unlimited iteration” — he refined a Submit button, added keyboard shortcuts, tweaked copy, adjusted prompts, and added confirmation messages. These workflow improvements “would have taken days or weeks of engineering and design back-and-forth” at his previous job.
The switch happened gradually. When he joined Jane Street in summer 2025, he used AI only for small UX papercut fixes. For bigger ideas, he still used Figma and docs. But “in the past 2 months the situations where I’ve reached for Figma have fallen off a cliff.” He now uses Claude for bigger work: “a half dozen other prototypes that make user-facing, data model, and library changes, including some that are 2000+ line diffs.” For some new apps, he even skips Figma entirely, “iterating on the visual design from the beginning with Claude.”
The one downside: reviewers receive a fully baked feature instead of a mockup, which can reduce design input. The team’s solution: write a reminder in the description that “prototypes are living proposal docs, the code is disposable, and a reviewer’s job is to give feedback about the design and user experience.” The reviewer eventually takes over the idea and implements production code in a separate feature.
📖 Read the full source: HN AI Agents
👀 See Also

Blackwell LLM Toolkit: NVFP4 Configs, Wheels, and Benchmarks for TensorRT-LLM on RTX Pro 6000
A community repo provides TensorRT-LLM configs, prebuilt LMCache wheels with sm_120 support, and benchmarks for Blackwell GPUs. Nemotron-3-Nano-Omni V3 hits 270 tok/s at 8k context on a single RTX Pro 6000.

EctoLedger: Open-source microVM sandbox for local AI agents with terminal access
EctoLedger is an open-source runtime firewall and ledger that provides microVM isolation for local AI agents with terminal access, running four prevention layers before executing commands in Apple Hypervisor.framework (macOS) or Firecracker microVM (Linux) environments.

Why Codex Still Beats Claude Code for Complex Python Monoliths
A senior developer compares Codex vs Claude Code on a production Python monolith with mixed architectural layers. Codex wins for back-end work due to better planning, code reuse, and harness-engineering adherence.

Claude Design vs Huashu-Design: A Head-to-Head on HTML Layouts and Rate Limits
Claude Design builds HTML prototypes fast but hits rate limits quickly. Huashu-Design, an open-source Claude Code skill, runs on the normal subscription with no separate rate limit—but takes 20 minutes vs 5.