ShaderPad launched this week as a 5.8kb library for dropping shaders onto websites without the usual graphics boilerplate. Creator Riley J. Shaw built it over years of wiring up the same scaffolding every time he wanted to put a shader sketch online. At 30x smaller than Three.js, it's aimed at artists and creative coders who just want an interactive graphic on their site without bundling a full 3D engine. He calls out Stripe specifically for using Three.js just to render a single fullscreen shader.
The library handles plumbing you'd rather not think about. It keeps work on the GPU instead of shuttling back to the CPU. It caches MediaPipe detection results across passes. Autosizing, save/share utilities, history buffers. React integration ready to go. The docs site at misery.co has interactive examples covering everything from basic uniforms to face tracking and segmentation.
What matters here for AI agent watchers is Shaw's actual development workflow. AI could probably generate the core functionality now, he says, but "the challenge for a project like this is not generating more code; it's deciding what to leave out." This resonates with the discussion in the AMD's ROCm article regarding AI agent assistance for code parity. AI tends to add features rather than maintain restraint. Where it shined was documentation. Shaw calls AI a "creative collaborator" that built thorough docs he then rewrote for clarity. The Lago CEO recently did the same with a custom Claude Skill to codify their writing style, finding that the emotional core stays human. He compares it to "a wide open version of Snapchat's new Imagine Lens."