Bryan Cantrill: Rust Can Catch Memory Errors, Not AI-Generated Complexity

Originally reported by InfoQ on November 16, 2026.

Rust’s ownership rules can help developers build safer infrastructure, but they cannot tell a team when an AI-generated system has become too complicated. That was the central warning in a talk by Bryan Cantrill, a DTrace co-creator and Oxide co-founder, which InfoQ edited into an article.

Bryan Cantrill: Rust Can Catch Memory Errors, Not AI-Generated Complexity (1)

Cantrill argued that tools designed to make software easier to write can shift costs into production. Garbage collection, for example, may spare developers from managing memory directly while introducing performance and latency concerns for some infrastructure workloads. Rust makes a different trade-off: developers must reason about ownership earlier, in exchange for stronger checks on memory use. He stressed that this does not make Rust easy to learn—or make code correct simply because it compiles.

Bryan Cantrill: Rust Can Catch Memory Errors, Not AI-Generated Complexity (2)

That compiler feedback can be useful when an LLM writes or revises Rust code, Cantrill said. But a model can also keep adding modules, abstractions and proposed improvements to a program that passes those checks. His concern is not just whether generated code runs, but whether anyone is keeping the system understandable.

Bryan Cantrill: Rust Can Catch Memory Errors, Not AI-Generated Complexity (3)

As an example, Cantrill cited developer Steve Yegge’s account of Gas Town, a Go-based software-development tool. Yegge said it worked well with Opus 4.6 and earlier, but that with Opus 4.7 it kept working on Gas Town itself instead of settling into the tasks it was meant to perform. That is Yegge’s account of the project, not a general finding about the model.

Bryan Cantrill: Rust Can Catch Memory Errors, Not AI-Generated Complexity (4)

Cantrill urged developers to use LLMs for clearly defined tasks and to understand the code they retain. He recommended learning Rust even when an LLM generates much of it: the developer still needs to read, debug and decide whether the software needs another change—or needs to stop changing.

Bryan Cantrill: Rust Can Catch Memory Errors, Not AI-Generated Complexity (5)

Source: infoq.cn (originally published 2026-11-16 14:28 UTC+8)