What's new in ui.sh – May 8, 2026

Summary

Key Points

• —The hardest challenge in building UI.sh has been figuring out the right way to integrate AI into user interface development, often requiring entirely new workflows rather than mimicking traditional design processes. 1:27
• —UI.sh incorporates design principles and implementation details to steer AI away from common mistakes (e.g., incorrect icon alignment, poor shadow implementation) and produce higher-quality, more professional UI code. 8:02
• —The team is shifting its strategy from a single, magical UI skill to identifying and documenting specific, useful AI-powered workflows, which will be surfaced through a redesigned UI.sh website. 10:31
• —A key discovery is the benefit of separating UI building into distinct phases: a rough ideation/design phase and a separate phase for making code production-ready and componentized, as AI struggles to do both simultaneously. 12:11
• —The redesigned UI.sh aims to be an educational hub, teaching users recommended processes, prompts, and sub-skills for specific tasks like drafting new UI, componentizing code, or converting sites to dark mode. 13:44
• —The team recognizes that different AI models excel at different tasks (e.g., OpenAI's new image model for UI design, Opus/Codex for code generation), necessitating workflows tailored to specific models rather than a single universal tool. 18:47
• —OpenAI's new GPT-2 image model is particularly effective for UI design, maintaining text consistency, and generating hyperrealistic dark mode versions of images, even from low-quality source photos. 20:50
• —The AI landscape is constantly changing, requiring UI.sh to continuously adapt, explore new capabilities, and update its documented workflows to provide ongoing value to users. 25:42
• —A core value proposition of UI.sh is for the team to stay on the bleeding edge of AI capabilities, discovering and surfacing new, non-obvious workflows (like generating high-quality images from poor sources) to customers. 29:35
• —The ultimate vision is for AI to function as an infinitely patient and skilled assistant, with humans remaining at the center, leveraging a toolkit of specialized AI capabilities rather than expecting a "magic wand" to complete entire projects. 40:27