Google research demonstrates how agentic LLMs can automatically generate Bug Reproduction Tests (BRTs) from vague bug reports, accelerating the debugging process.

• 75% success rate at bug reproduction for real-world internal bug reports
• 39% cost reduction in time required to generate BRTs compared to developers
• Agentic approach enables autonomous exploration of systems to find reproduction paths
• Successfully integrated into Google's engineering workflow for production use

This research addresses a critical engineering challenge: most bug reports lack sufficient detail to reproduce issues efficiently. By automating the bug reproduction process, development teams can significantly reduce time-to-repair and improve software reliability.

Agentic Bug Reproduction for Effective Automated Program Repair at Google