Using AI to Delete Code: Agentic Tech Debt Cleanup at Uber (www.uber.com)

Uber has developed an AI system called AutoClean to automatically delete stale feature flags from its codebase, addressing the technical debt generated by unused code. This system significantly enhances the software development lifecycle by continuously identifying and cleaning up legacy flags, which, if neglected, can lead to complications and inefficiencies for developers and AI agents alike. AutoClean has successfully cleaned over 16,000 flags and 1,000,000 lines of code, ensuring that Uber's vast and complex codebase remains maintainable and efficient. The significance of this innovation lies in its ability to streamline code maintenance at scale, where manual cleanup becomes time-consuming and often deprioritized. AutoClean employs a combination of AI-driven agents and deterministic tools to validate changes automatically before human review, improving safety and effectiveness in making large-scale code updates. By focusing on key properties such as preserving program semantics and avoiding unintended artifacts, AutoClean enables engineers to concentrate on high-level decisions rather than low-level code correctness, thereby enhancing overall developer productivity.