AI now writes Rust and Go well enough that the performance-vs-ergonomics tradeoff that kept teams on Python and TypeScript has collapsed.
Key Takeaways
Strong type systems and fast compile-check loops give agents tighter iteration cycles; Rust’s compiler feedback accidentally optimized it for AI-assisted development.
Concrete 2026 data points: Anthropic researcher built a C compiler in Rust (100k lines, ~$20k); Ladybird’s JS engine ported C++ to Rust in two weeks with zero regressions across 65k+ tests.
Microsoft shipped TypeScript 7.0 beta in Go, roughly 10x faster than 6.0; OpenAI acquired Astral (uv saves Codex ~1M compute minutes/week); Anthropic acquired Bun.
Python’s ecosystem is quietly becoming a Rust ecosystem: Pydantic core, Polars, Hugging Face tokenizers, orjson, ruff, and uv are all Rust under the hood.
Armin Ronacher ported MiniJinja to Go in 45 minutes of human time and $60 of API cost, shifting the contribution unit from patch to port.
Hacker News Comment Review
Commenters are split: Python still dominates AI/ML training data and tooling, making LLM output fast and readable, but dynamic typing means broken code paths can silently survive even passing test suites.
Several commenters argue the ideal AI-coding language needs type safety for guardrails, concise syntax for reviewability, and a large public corpus – no single language fully satisfies all three today.
A practical counterpoint: use the language you know best, since the human’s job is now reviewing and judging AI output, not writing it – unfamiliar languages shift risk to the review phase.
Notable Comments
@kylec: shipped a Go web service without knowing Go beforehand; cites native threading and simple container deployment as practical wins.
@dakiol: “The LLM could have broken some path that only gets run in production” – strongly-typed compilers catch this; non-negotiable in large codebases.