I Was Wrong About Rust...
GitHub documented a Rust rewrite of its Copilot runtime, revealing that AI agents handled most coding tasks across 128 incremental pull requests. A single developer completed the project in months, whereas a team previously required a year or two. Analysis of compiler errors showed that 84 percent were standard type errors, while only 1.7 percent involved Rust-specific borrowing or lifetimes. The speaker admits he was previously wrong about the compiler being the primary reason AI performs well with Rust. Success relied on end-to-end tests and a second agent checking the first, rather than compiler features alone.
break your ai's check on purpose, once
A factory verifies quality control by sending a known broken toaster down the line to ensure the station catches it. One should test an AI check immediately after writing it by intentionally changing one thing to make the page wrong.