@dhh
Post
Had GPT Sol do an optimization run on this. It immediately made it 63% faster. So now ttfx is 14x faster than the original tte. https://github.com/omacom-io/ttfx/pull/2
Quoted post by DHH (@dhh) Fable one-shotted a Rust rewrite of the TerminalTextEffects Python library in 11M tokens. Startup time went from 87ms to 2ms and rendering speed is up by 9.6x. Now zero dependencies and a 3mb single exec 🤯 https://claude.ai/code/artifact/287825bc-7aad-4541-a7e7-fa4ba8d03612
Explanation
What it says DHH says a Rust rewrite of the Python TerminalTextEffects library was generated in one shot by “Fable” using 11M tokens. He reports startup dropping from 87 ms to 2 ms, rendering becoming 9.6× faster, dependencies going to zero, and the result becoming a ~3 MB standalone executable. A subsequent “GPT Sol” optimization pass allegedly made that Rust version another 63% faster, yielding ~14× the original rendering performance.
Context This is essentially a claim about AI-assisted software reimplementation and optimization: first translating/rebuilding a Python library in Rust, then doing a second optimization pass. The post links to the optimization PR, but the saved material does not include benchmarks, methodology, code review, correctness tests, hardware, workload characteristics, or what “11M tokens” specifically measures.
Why it matters The interesting part is not merely “AI writes code,” but the suggested workflow: use a model to perform a wholesale language/runtime change, then use another optimization pass to extract additional performance. If reproducible without regressions, that materially lowers the cost of aggressive rewrites. The missing question is correctness: speedups are only meaningful if behavior, visual output, portability, and benchmark methodology remain equivalent.