Tool Reviews

Paul Ford Argues That the Web Is More Than Sufficient

Claude Opus 5.5 ported a Three.js Faberge egg to WebGL, showing AI can move from libraries to raw APIs, though size and details may differ.

Paul Ford Argues That the Web Is More Than Sufficient

Key Takeaways

  • A Hacker News commenter used Claude Opus 5.5 in an existing chat thread to port a 77KB Three.js “nudibranch Faberge egg” project into a 98KB Vanilla JS/WebGL page, showing AI can translate high-level 3D library code into lower-level browser APIs.
  • The experiment is not a performance or quality benchmark: the file-size difference is not explained, the outputs are not identical, and the port is a broad capability demonstration rather than a pixel-perfect rewrite.
  • For creative coders, the result matters most visually; the commenter found Three.js code slightly easier to scan but rarely reads generated code closely, so library-free output can be acceptable if it looks close to the original.
  • The broader takeaway is that AI coding assistants may lower the barrier for non-experts to try WebGL-style work, while raising questions for developers about generated code size, readability, and maintainability.

Meta description: Claude Opus 5.5 ported a 77KB Three.js nudibranch Faberge egg to a 98KB Vanilla JS WebGL page, showing AI can move from libraries to raw APIs.

Paul Ford Argues That the Web Is More Than Sufficient

Tags: hacker-news, javascript, ai, claude, three-js

Featured image: Abstract friendly illustration of a nudibranch inside a jeweled egg made of wireframe 3D geometry.

A Hacker News comment on Paul Ford's essay The Web Is More Than Enough prompted a small browser experiment that put Claude Opus 5.5 to work on 3D web code. The test started with a Nudibranch-themed Faberge Egg made in Three.js, shared as a 77KB page. The commenter then asked Claude, specifically Opus 5.5 Medium running in regular Claude Chat, to port the same project to Vanilla JS and WebGL. The ported page came out at 98KB.

The context: a chat thread, not a fresh prompt

The request was made in the same thread where the project had originally been built. That detail keeps the experiment inside a familiar AI coding workflow: the model had the earlier Three.js work in context, then was asked to translate it into lower-level browser code. The result kept the creative idea—a nudibranch inside a Faberge Egg—while removing the dependency on Three.js.

What changed between 77KB and 98KB

Three.js is a JavaScript library that wraps WebGL, the browser API used to draw 3D graphics, in a simpler scene-and-object model. Vanilla JS means writing browser code without that library. The experiment does not turn the file sizes into a performance claim. It only reports that the Three.js version was 77KB and the Vanilla JS version was 98KB. The source does not explain the extra size, and the comparison should not be read as a benchmark of speed, quality, or maintainability.

The outputs were not identical

The two pages are not identical. A shared Claude transcript, https://claude.ai/share/e56d163e-2793-4677-aa11-191d5058b414, contains the model's own description of the differences. The commenter says they did not fully understand why the outputs differed. That caveat matters: the experiment shows a successful port in broad form, not a pixel-perfect rewrite of every detail.

Why the code difference may not matter

For the person who ran the test, reading the code was not a major issue. They said they found the Three.js version slightly easier to understand by eyeballing it, but added that they rarely look at the code for these 3D projects, so the difference was not a big deal in daily use. That is a useful practical note for creative coding: library-backed code may be easier to scan, while lower-level generated code may be accepted if the visual result works and the source is not reviewed often.

What readers can take from it

The broader takeaway is that an AI coding assistant can move a small creative coding project from a high-level 3D library down to lower-level browser graphics code. For non-experts, that lowers the barrier to trying WebGL-style work without first mastering every Three.js abstraction. For developers, it raises a practical question: generated Vanilla JS may be larger and harder to read than a library-backed version, even if it runs and looks close to the original. The experiment is a small proof of capability, not a claim that Three.js is unnecessary.

Comments (0)

  1. No comments yet. Be the first to share what worked for you.

Leave a comment

Comments are reviewed before they appear. Your email address is not published.