DoorDash Uses Multi Agent LLMs to Clean up 60,000 Feature Flags
Not a pilot. DoorDash deployed multi-agent LLMs across 60,000 feature flags — 90% success rate, $4.79 per cleanup.
Why it matters
DoorDash's production agent system demonstrates practical autonomous workflow automation at enterprise scale: agents handling code review, validation, and git operations with human-in-the-loop approval. This is a credible case study for practitioners evaluating agents for internal tooling and technical debt reduction.
The key facts
7 to know60,000+ feature flags across 623 repositories
90% success rate (45 of 50 flags tested produced usable PRs)
13.8 minutes average cleanup time per flag
$4.79 average cost per cleanup
Multi-agent workflow with live experimentation data integration (MCP)
Human approval gate + isolated Git worktrees + parallel agent execution
Automated validation step
Go to the source
InfoQ AI/MLinfoq.com
Publisher excerpt: DoorDash built a multi-agent LLM system to automate stale feature flag cleanup across more than 60,000 flags and 623 repositories. The workflow combines live experimentation data through MCP, engineer approval, isolated Git worktrees, parallel agents, and automated validation. In an evaluation of…