AI as Compiler: The Concept That Turns Pseudocode Into Production Code

A developer has proposed a conceptual programming language where an AI large language model acts as the compiler, synthesizing undefined functions from context and converting loosely written pseudocode into production-ready code in languages like Rust or C++. In this model, the LLM resolves dependencies, auto-generates missing functions, and hands the output to traditional compilers like rustc or clang to produce optimized native binaries. To reduce hallucinations, the system would run generated code in a sandbox, auto-write unit tests, and self-correct based on failure traces until all tests pass. However, the concept faces two significant hurdles: non-determinism across compilation runs and the latency of LLM inference, which could disrupt developer workflows. The author acknowledges these as serious limitations today but suggests shrinking, faster local models could make this a viable approach in the future.