Microsoft Open‑Sources Code‑Testing Generator to Vet AI‑Written Code
Microsoft released an open‑source "code‑testing‑generator" agent that scans a repository, writes polyglot unit tests, runs them, and only then marks a task complete. In internal benchmarks it hit 92.1% task completion versus 78.9% for GitHub Copilot, closing the trust gap for LLM‑generated code.