Rust is unsurprisingly the largest beneficiary (among programming languages) of coding agents. It is notoriously difficult to master due to the large API surface and strictness; as a result mainstream devs didn't use it much. The benefits are huge though. Super-fast like C/C++, but without the safety issues. If it compiles it is probably good -- a built-in automatic verifier that works well with AI. OpenAI/Anthropic should be using it more for their sandboxes :) At Google we had a super-long and complex C++ style guide that encoded years of tribal knowledge on how to use C++ well and good libraries that made it safer and stricter (a lot of them by Jeff and Sanjay actually). If you didn't abide you were rightly yelled at during code review. A lot good conventions are just built-in to Rust, and the style guide could be reduced to just don't use 'unsafe'. The main argument against Rust was lower dev velocity, and on the opposite end is fast and loose Python (which researchers love because they're often weak at coding). But coding agents turn this on its head. AI knows Rust well, gets immediate feedback on hallucinations or bad code, and can self-correct. It perfectly complements the strengths of LLMs.
While Everyone is Waiting for the Next Model, Your Agent Can Learn Tonight