Writing Rust code that's fast by asking agents to make the code faster

A researcher tested whether iterative prompting can get LLMs to write significantly faster Rust code, finding that modern agentic LLMs can achieve 2x-20x speedups depending on the domain when given appropriate constraints and guardrails. The author used Rust as the target language due to its speed and Python integration capabilities via PyO3, avoiding unsafe code to maintain memory safety while optimizing performance. The experiment began with reimplementing machine learning algorithms like UMAP in Rust to validate whether LLMs could improve upon battle-tested C implementations.