Simon Willison tested Claude Opus 5.5 with a 3D sea-anemone-egg artwork: he first built it with Three.js, then asked the model to rewrite it in the same conversation as native JavaScript and WebGL. The generated result ran successfully, although the two versions were not completely identical. This case matters because code generation is no longer limited to writing a few lines of functionality; it is approaching the point where AI can independently deliver a small web project.
What this is
Three.js is a JavaScript framework for creating 3D graphics on webpages, while WebGL lets browsers process graphics directly. Willison's approach was to have the model convert framework-dependent code into a standalone version. After the rewrite, the file grew from roughly 77KB to 98KB, while still producing a similar 3D effect. He also acknowledged that he understands the original code better and has not fully identified why the versions differ.
Industry view
This shows that AI coding tools are moving beyond the stage of generating runnable snippets and beginning to take on more complete tasks such as translation, refactoring, and migration. For ordinary development teams, this could shorten the distance between an idea and a working prototype. It also means more non-specialists can use natural language to create interactive content.
The risks are equally clear: runnable does not mean correct. Visual deviations, code-logic problems, and maintenance costs may all surface only after deployment. The model's explanations of why differences emerged should not be accepted as definitive answers. Without testing, version control, and human review, this capability is better suited to accelerating prototypes than modifying production systems without supervision.
Impact on regular people
For enterprise IT: Teams can validate small tools, interactive pages, and 3D prototypes faster, but existing system upgrades still require testing, access controls, and human review. Governance should not be relaxed simply because a model can produce runnable code.
For individual careers: Product, operations, and training professionals can directly create interactive presentation materials. The main value lies in shortening communication and trial-and-error cycles, not replacing professional developers.
For the consumer market: Personalized interactive content and lightweight webpages may increase in the near term. Yet users will have difficulty judging code quality from a page's appearance alone, while privacy, security, and long-term maintenance will still depend on whoever made it.