Agentic Proof-Oriented Programming (risemsr.github.io)

Researchers at Microsoft Research are pioneering a new approach called Agentic Proof-Oriented Programming (PoP), which leverages AI models like Copilot CLI to assist developers in creating programs with formal proofs of correctness. This innovative method allows human experts to concentrate on high-level design and specifications while AI handles the intricate details of generating formal proofs. The implications of this technology could drastically increase the scalability of proof-oriented programming, enabling small teams to manage larger verified systems with greater efficiency. In practice, the author successfully utilized Copilot CLI to generate 10,000 lines of formally verified code in Pulse—a language embedded in F* for proof-oriented programming. Examples include implementing and verifying complex structures like a bubble sort, a priority queue, and concurrency controls like reader-writer locks, tasks that would typically require extensive manual work. The use of AI not only accelerates code production but also enhances the accuracy of the proofs, making it easier for developers to review and integrate code into production systems. As this research unfolds, it promises to stimulate further exploration and innovation in the realm of AI-assisted programming.