Automating Feature Flag Cleanup at Scale with Multi-Agent LLM Systems
What it is
This paper describes an industrial implementation at DoorDash of a multi-agent LLM system designed to automatically identify and remove stale feature flags. The system integrates live rollout data via MCP, processes Jira tickets, and generates pull requests for code cleanup, achieving a 90% approval rate in evaluation.
The problem it solves
Manual feature flag cleanup is time-consuming (1-2 hours per flag) and existing rule-based tools are insufficient for complex codebases, leading to accumulated technical debt and operational risk.
What you could use it for
- Automating the removal of stale feature flags in large-scale monorepos
- Reducing technical debt associated with dormant code paths
- Generating pull requests for code cleanup based on live feature rollout status
Why it matters
Significantly reduces the manual engineering effort required to maintain codebases by automating a repetitive and error-prone cleanup task, potentially saving thousands of hours across large organizations.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.