Microsoft Research releases PDD Kit to formally verify agent-written full-stack web apps
PDD Kit converts developer intent into formal specifications, proves AI-generated code meets them, and falls back to tests where proofs are impractical.
1 source · cross-referenced
- PDD Kit formalizes developer intent into specifications and proves AI-generated code meets them for all inputs.
- The toolkit was used to formally verify login logic in a Rust-based full-stack web app within GitHub Next’s Chopin environment.
- PDD Kit automates proof generation and high-quality test creation without requiring formal methods expertise.
- The approach aims to scale code review as AI agents generate increasingly large codebases.
Microsoft Research released PDD Kit, a toolkit designed to formally verify code generated by AI agents, addressing the challenge of scaling human review for large codebases. The toolkit converts developer intent into formal specifications in Verus, a proof language for Rust co-developed by Microsoft Research and academic partners, and proves the generated code meets those specifications for every possible input.
PDD Kit was applied to verify the login logic of Pet Store, a Rust-based full-stack web app, within GitHub Next’s Chopin environment. The toolkit uncovered existing bugs and proved the absence of login bugs relative to the specification, all while developers reviewed intent rather than individual lines of code.
The workflow in PDD Kit involves two halves: first, team members review and record intent in plain language, with PDD Kit guiding the process and explaining relevant code. Parts of the intent that can be formalized are converted into specifications, while the rest are handled via informal specifications for testing. PDD Kit then automatically generates proofs for formal specifications and high-quality tests for informal ones.
The toolkit is positioned as a shift toward Proof-Driven Development (PDD), where formal methods are made accessible to engineers without specialized backgrounds. PDD Kit runs as a partner inside Chopin, GitHub Next’s real-time, multiplayer planning environment, enabling collaboration between teams and agents in a shared document.
PDD Kit distinguishes between formal verification—mathematical proofs that code satisfies specifications for all inputs—and empirical verification, which relies on tests for scenarios where proofs are impractical, such as third-party libraries or user interfaces. Formal verification in PDD Kit is performed using Verus, while empirical verification uses unit and end-to-end tests, terminal automation, browser agents, and computer-use agents.
- Nov 11, 2026 · The Verge — AI
Google DeepMind unveils AlphaGenome 2 to predict genetic variant effects on cell types
Trust79 - Nov 11, 2026 · Google DeepMind — Blog
Google DeepMind unveils AlphaGenome 2 to predict DNA variant effects in specific cell types
Trust84 - Nov 11, 2026 · arXiv cs.AI
Researchers demonstrate auditable human-agent teams for autonomous synchrotron X-ray experiments
Trust79