Can Rust Make Unsafe AI Agent Actions Unrepresentable?

In a recent article on Dev.to, developer Ken Walger explores how the Rust programming language can be used to enforce AI agent safety through its type system. Rather than relying on runtime guardrails that developers must remember to implement, Walger proposes using the 'typestate' pattern to make unsafe actions unrepresentable at compile time. By defining distinct types for different states—such as 'ProposedWrite' and 'AdmittedWrite'—and restricting access to constructors, developers can ensure that only validated actions reach critical functions like persistence. This approach shifts the burden of safety from human memory to the compiler, preventing invalid states from ever reaching production. While the author notes that this does not solve all AI safety challenges, such as policy correctness or prompt injection, it provides a robust architectural framework for preventing common, architecturally invalid transitions in agentic systems.