Summary
Y Combinator hosts Steven Haney, founder of Paper, an AI-native design tool built from the ground up for the agent era. Unlike Figma and other established tools that rely on custom rendering engines, Paper uses HTML and CSS as its core rendering layer — a deliberate technical bet that pays off when agents interact with design files, since models natively understand web standards and can work with Paper at lower token cost, faster speeds, and higher accuracy than proprietary formats.
Haney demos the tool live, showing how agents can generate multiple layout variations, branch design directions in parallel, and iterate on visual components through plain-language instructions. He shares specific model preferences: Fable for pure visual taste (though slow, making it impractical for demos), Cursor Composer 2.5 for precision and speed, and Anthropic’s Claude models for systematic large-batch tasks like processing 100 artboards in sequence. He also offers practical heuristics for steering AI output away from “model slop” — pulling back the bold font weights models default to, constraining the number of distinct font sizes, and using branching to explore creative directions before committing.
The broader argument Haney makes, with YC’s implicit endorsement, is that design remains a powerful competitive differentiator precisely because generic AI output risks homogenizing product aesthetics. Paper is positioned as the tool that lets human designers move fast with agents while preserving the craft judgment that makes great products stand out.
📺 Source: Y Combinator · Published August 07, 2026
🏷️ Format: Interview
