Language Agnostic Programming: Why you may still need code (joaquimrocha.com)

Author Joaquim Rocha argues that while LLMs and copilots can generate working code from natural language, programming languages will remain essential—chiefly for debugging and unambiguous specification. Natural language is inherently ambiguous (the milk/eggs joke), whereas code precisely encodes semantics such as control flow, types, memory and concurrency behavior. That precision matters when programs fail: developers need readable, traceable artifacts to reason about edge cases, undefined behaviors, and correctness—roles LLMs can assist with but not fully replace. Rocha proposes a “language-agnostic programming” workflow where AI generates a single precise implementation (e.g., Rust for its ownership and type guarantees) and LLMs present, edit, or translate that program into whatever surface language a developer prefers (Python, Go, TypeScript, etc.). Benefits include easier onboarding, cross-team collaboration, and democratization of code comprehension. Key technical hurdles are real: mapping Rust’s ownership model, C++ templates, or language-specific idioms and libraries into higher-level views without losing semantics; maintaining bidirectional consistency when edits are made in different languages; and requiring very fast, highly accurate model-based translation tools. The idea is speculative but actionable: current tools can convert snippets today, and improved LLM speed, correctness, and tooling could make language-agnostic debugging a practical future.