Descriptions:
Daniel Szoke, Rust SDK maintainer at Sentry, makes a counterintuitive case at AI Engineer 2026: Rust — not Python or TypeScript — is the ideal language for AI-assisted and agentic coding. The conventional wisdom points to Python and TypeScript because LLMs write them fluently and they’re fast to scaffold. Szoke argues that this framing optimizes for the wrong thing.
The core of his argument is that LLMs are non-deterministic by design and will always produce mistakes. In dynamic languages like JavaScript and Python, those mistakes can silently propagate — a well-formatted function with a subtle logical error passes review but fails in production. Rust’s strict compiler, by contrast, enforces type safety, memory safety, and concurrency invariants at compile time: if the code compiles, an entire class of bugs is provably absent. That deterministic guardrail, Szoke argues, is exactly what you want wrapping a fallible generative system.
He also draws on Yuval Noah Harari’s concept of LLMs as “alien intelligence” — systems whose failure modes are structurally different from human error — to motivate why tests and code review agents alone are insufficient. The talk is aimed at teams choosing a language stack for new agent-first projects and challenges the assumption that ease-of-generation is the most important selection criterion.
📺 Source: AI Engineer · Published May 27, 2026
🏷️ Format: Opinion Editorial
