Hamilton Greene has spent years bouncing between F#, TypeScript, and C#. Always settling. Each language falls short somewhere. The ecosystem isn't there. The types don't hold up. The boilerplate becomes unbearable. Then someone at Recurse Center mentioned that Rust is worth using just for the types. That clicked. What had always stopped him was the learning curve. Ownership, borrowing, lifetimes. Systems programmer territory. But with AI coding assistants now handling standard code generation, he wondered if that curve had flattened enough for application developers.
His answer: "High-Level Rust." The approach is straightforward. Use enums and structs to model your domain. Write functional-style logic with immutable data and pure functions. Clone generously instead of wrestling with the borrow checker. This mirrors the approach proposed in a recent analysis of how functional programming fixes AI agent failures, emphasizing side effects at the edge and explicit data flow. Accept a 10-30% performance hit as the cost of doing business. For web APIs, CRUD services, and business logic where correctness beats raw speed, this trade makes sense. Rust is fast enough at baseline that losing some performance still leaves you ahead of most compiled languages.
The strategy also matches what AI assistants are good at. Models like GPT-4 and Claude 3.5 Sonnet can generate Rust syntax and boilerplate all day. What they can't do is reason about ownership graphs or manage complex lifetimes. Ask an LLM to write code with multiple mutable references or fiddly borrowing patterns. You'll get something that won't compile. But ask it to write functional-style Rust with simple clones? Success rates climb. One developer, Lalit Maganti, spent eight years building SQLite tools before using AI to generate a Rust rewrite, illustrating how AI helps but requires developer oversight. Developers can use Rust without becoming systems experts, without waiting for LLMs to get better at memory management puzzles.