The company says it manages thousands of flags across mobile applications and backend services. Once a flag has fully rolled out, its conditional branches, configuration, parameters, and tests can remain as technical debt.
AutoClean connects to Uber’s feature-flag system of record, identifies flags considered safe to remove, rewrites code, addresses CI failures, validates the resulting diff, and routes it to the owning team for review. It runs continuously and keeps review-ready changes fresh until a human reviewer makes the final decision.
The core agent has access to the codebase plus a scoped tool set for file search and editing. For boolean algebra and conditional simplification, Uber uses PolyglotPiranha, its code-transformation engine, exposed to the agent through MCP rather than relying on the model to make those rewrites itself.
The post does not provide comparative measurements for cost, acceptance rate, defect rate, or the number of changes merged by AutoClean. It also does not describe a public release or external access to the system.