AI demos are entering a strange era:
half technical benchmark,
half magic trick,
half “wait why do I suddenly want to try this?”
Claude Code Fable 5 writing Three.js — Is it really that magical? My hands-on test conclusion might differ from what you expect.
I actually ran the full test with Fable 5. Using a single prompt to generate Three.js code does produce nice results — the visuals are pretty.
But here’s the thing: my partner and I already achieved this exact kind of thing two months ago.
Back at the end of March, we used one prompt to generate a “forked-path garden coffee cup” in Three.js. At the time we were combining Codex Claude Code local LLMs, and the output even included butterflies.
Honestly, Fable 5 isn’t some god-tier breakthrough — because this was already possible.
When a model company keeps pouring so much effort into design and 3D visuals, it reveals a somewhat harsh truth: the ceiling for AI coding is getting close.
You can already smell it from Claude’s “Design” features. Pure logic and functionality have been squeezed pretty dry. What’s left is endless polishing of details — and that kind of precision aesthetic work still fundamentally requires human in the loop.
In plain terms, they’re packaging the creative and aesthetic work that humans should be doing, dressing it up as AI’s new magic trick, and then selling you a more expensive plan.This is arbitrage and narrative marketing. It’s not a miraculous leap in the model itself. Don’t get too excited.
The comparison videos are in the comments.