Microsoft Research releases PDD Kit to formally verify agent generated code

Microsoft Research and Microsoft reported on November 5, 2026, that they have introduced PDD Kit, a framework to formally verify full-stack web applications written by artificial intelligence agents. The system aims to address security and correctness challenges created when coding agents generate tens or hundreds of thousands of lines of code in single sessions. [1][2]

According to Microsoft Research, coding agents now create software at a scale that makes line-by-line human review impractical. Instead of reviewing every line in a large pull request, developers using PDD Kit focus on reviewing the core intent of the application. [1]

Microsoft Research explained that PDD Kit functions by converting developer intent into a Verus specification. The system then mathematically proves that the AI-generated code meets the specification for every possible input, while relying on standard testing methods where formal mathematical proofs are not practical. [1]

Microsoft indicated that software development teams are adapting to agent-generated code by shifting their focus toward investing in high-level planning and relevant testing frameworks rather than manual line reviews. [1][2]