Bend

AI-generated code now has mathematically provable guarantees against violating specified laws, eliminating entire classes of bugs.

Bend, a new programming language, has been released, designed to prevent AI-generated code mistakes through formal proofs and high-speed compilation. The language aims to provide an unambiguous method for instructing AI systems, especially in a post-AGI economy where humans may cease direct coding. Bend compiles to native code, achieving near C-level speed on a single core and significant speedups on multi-core processors and GPUs through automatic parallelism. Its type checker also functions as a proof checker, designed to be fast, allowing AI agents to verify code changes rapidly. A key feature is `LAWS.bend`, which defines rules that AI cannot violate, preventing bugs by making them mathematically impossible to merge. The installation involves a simple curl command, and users are instructed to add Bend to their `AGENTS.md` file to enable its use with AI agents. The system is designed for backend development on Linux and macOS.

Bend addresses the critical challenge of AI-generated code reliability by integrating formal verification directly into the development workflow. The language's emphasis on speed, both in compilation and execution, alongside its unique proof-checking type system, positions it as a potential solution for large-scale AI-driven software development. By making it impossible for AI to ship code that violates declared 'laws,' Bend aims to shift the paradigm from debugging to provable correctness, which could significantly reduce development cycles and enhance software stability.

The introduction of Bend signals a maturing market for AI development tools, moving beyond mere code generation to rigorous validation. Companies and developers relying on AI for coding assistance will be able to build more complex and critical applications with greater confidence. The ability to enforce 'laws' through mathematical proof means that the cost of AI errors, particularly in sensitive applications, could be drastically reduced, impacting industries where reliability is paramount.