Vera: a programming language designed for machines to write (github.com)

Vera is a newly launched programming language crafted specifically for large language models (LLMs) to generate code more effectively. Its design eliminates traditional variable names in favor of structural references, where parameters are specified by their position (@Int.0, @Int.1). This transformation aims to enhance code clarity and verification, crucial for LLMs that typically struggle with coherence over larger codebases. Vera enforces strict contracts for function signatures, mandating preconditions, postconditions, and effect declarations, which the compiler statically checks, ensuring reliability and reducing common naming errors that can lead to bugs. The significance of Vera lies in its potential to fundamentally shift how LLMs generate code, addressing their limitations in maintaining logical consistency and state across applications. By making explicit what needs to be checked and verified, Vera allows LLMs to focus on correctness rather than syntax. The language compiles to WebAssembly, enabling execution both in command-line environments and web browsers, thus broadening its applicability. As LLMs increasingly become authors of code, Vera sets a precedent for future programming languages tailored to machine learning, potentially leading to more robust software development practices driven by AI capabilities.