Brian built Notion’s internal 'prototype playground': a collaborative Next.js app where designers rapidly build, test, and share functional prototypes—moving beyond static Figma mockups. This environment enables early exposure to technical reality, surfacing design flaws before engineering handoff. He demonstrates using Claude Code’s plan mode, custom slash commands (e.g., '/create prototype', '/figma'), and skills like 'find icon' to automate repetitive tasks and improve fidelity. A key insight is teaching Claude to self-correct—running ESLint, validating outputs, or deploying via GitHub—rather than manually fixing errors. The workflow converts Figma designs into ~80% production-ready UIs and supports rapid iteration on AI-native features, which demand real-time interaction modeling impossible in static tools. Brian emphasizes that for AI-powered products, designers must engage directly with code to understand latency, variability, and model behavior—shifting design from visual speculation to grounded, executable experimentation. His tool stack centers on Claude Code for quality, Cursor for debugging, and Opus 4.5 for structured planning—all guided by the principle: when Claude asks you to do something, teach it to do that itself.