Skip to content- Sources: primary
- Summary: Cantrill contrasts language features that eased development at the operator's expense, such as garbage collection, dynamic typing, exceptions, ORMs and RPC, with Rust, which removed garbage collection and green threads and traded a steeper learning curve for safety. He argues that this learning curve becomes a strength with LLMs, because code that compiles says something about its correctness, while LLMs lack a programmer's laziness, keep finding more work, and burned down the Gas Town software factory. He recommends using LLMs heavily for code comprehension, writing comments yourself as a check on understanding, and points to Oxide's public RFD 576 on LLM use.
- Why it matters: It gives a concrete rule set for agent use in Rust codebases: lean on agents for comprehension and well-defined tasks, keep the abstraction design and comments human.
- Background: Rust's ownership model gives each value a single owner and checks borrows at compile time, which removes the need for a garbage collector while preventing use-after-free and data races. Cantrill uses the example that Rust's standard BTreeMap is practical because the compiler knows who holds references to the values the tree moves, while a C B-tree library would need a complex callback contract. Accidental versus essential complexity is Fred Brooks's distinction between complexity created by past choices and complexity inherent in the problem.