Bend pitches a fast proof-checked programming language for AI agents, with native CPU and GPU parallelism

Bend presents itself as a programming language designed to let AI agents generate code while enforcing developer-defined rules through proofs. It claims near-C performance on one core, automatic parallelism across CPUs and GPUs, fast proof checking, and a Python-like syntax for building vibe-coded apps with fewer mistakes.

The project reflects a growing push toward verifiable AI coding, where agents are constrained by formal specifications rather than reviewed line by line by humans. If its performance and proof-checking claims hold up in real-world projects, Bend could appeal to teams trying to combine AI-assisted development with stronger correctness guarantees.