Claire Vo gives an early, hands-on account of Fable 5.2 on How I AI. She previews selected artifacts, not the complete comparative benchmark she says she plans to run later.
Where it helped her
- From a tweet about software factories and a company brand book, she iterated on an animated explainer. She especially likes how the model animated static mascot images into moving sprites.
- She says it generated usable assembly diagrams and ordered steps from blurry pictures of an unboxed furniture kit. She followed the instructions with an assembler; this is one personal case, not a validated geometry benchmark.
- A brand-themed bug-defense game supplies the playful example: protecting production against bugs, adding agents and tests, and facing a giant monorepo as a final enemy. She says she intended to publish it, not that its final release was verified in this video.
- She reports progress on a messy ChatPRD data-pipeline reorganization, but says she was only halfway through. That is not a completed migration or a measured improvement.
What didn’t work and what took iteration
- When given little visual direction, the model made a browser homepage and storybook that Vo found ugly. She suggests using another model for front-end design while keeping Fable for other tasks.
- Her 3D writing environment demonstrates a typewriter, alternate scenes, music and rain, timers, handwriting and a dot-matrix-printer view. She calls the original typewriter a single-shot starting point, then explicitly says she tweaked it and kept adding features. The whole environment did not emerge in one prompt.
- Vo calls the model fast and previews timings for tasks in her own bench, but promises a fuller multi-model comparison later. Claims about scientific problems, general benchmark leadership, pricing and availability in the introduction are not independently tested by these demos.
“I’ve done some tweaks. I’ve kept giving it new features.” — Claire Vo, on the typewriter