June 30, 2026
Rust never sleeps, Julia fans do
From Julia to Rust: a differentiable tensor stack for scientific computing
Scientists dump Julia for Rust, and the comments instantly turn messy
TLDR: The team behind tenferro-rs says they moved a major science-computing project from Julia to Rust because bigger codebases felt harder to trust and maintain. Commenters are split between excitement over a long-awaited Rust tool, suspicion about the AI-written sales-pitch tone, and jokes that Rust is only fun when a bot writes it.
A new project called tenferro-rs just landed with a big promise: give scientists a Rust-based tool for heavy math and AI work, after years of building similar systems in Julia. The team’s case is simple but spicy: Julia was lovely for fast experiments, but as projects grew, it became harder to keep things reliable and fast to develop. Their answer? Move the engine to Rust and glue together missing pieces into one stack for number-crunching, automatic gradients, and GPU support.
But the real show was in the reactions. One former Julia user basically shrugged and said, yes, fair points, but also admitted the quiet part out loud: Julia is fun, and Rust only starts to feel tolerable when an AI is doing the typing for you. That line alone feels destined for comment-thread hall of fame. Another commenter was far less amused, calling out the reveal that the post was AI-generated and side-eyeing the whole thing as giving off strong "use my project" sales pitch energy. Translation for non-tech readers: some people loved the software idea, but hated the vibe.
Meanwhile, the hype squad showed up too. One commenter sounded genuinely thrilled that Rust might finally get a JAX-like tool scientists have been craving forever. Then came the practical grumbling: support for Apple devices seems awkward, and extending the system could be painful. So the mood is deliciously split: big excitement, big skepticism, and at least one excellent joke about outsourcing Rust suffering to robots.
Key Points
- •The article announces tenferro-rs, a Rust-native dense tensor stack whose first crates were released on crates.io on June 23, 2026 (JST).
- •tenferro-rs is described as providing linear algebra, eager autodiff, traced transforms, einsum, FFT, extensible operations, and explicit CPU/CUDA backends.
- •The tensor4all team's earlier tensor-network work was largely done in Julia, but the article says larger Julia codebases created issues with runtime type instability, compilation time, and correctness checking.
- •The article argues that Rust had strong component libraries such as faer, CubeCL, ndarray, Burn, candle, and numr, but lacked a unifying scientific-computing tensor layer with autodiff through einsum.
- •The post says the shift toward AI-generated code changes language tradeoffs, making stronger correctness and systems guarantees more important in the author's view.