● Covered by 1 source · 1 reportMedium impact1 neutral
DoorDash Deploys Multi-Agent LLM System to Automate Feature Flag Cleanup
Key points
- DoorDash's LLM system automates stale feature flag cleanup.
- The system produced usable pull requests for 45 out of 50 flags.
- Cleanup time averaged 13.8 minutes and $4.79 per flag.
- DoorDash manages over 60,000 feature flags across 623 repositories.
Automating Feature Flag Management
DoorDash has implemented a multi-agent LLM system to streamline the process of identifying and removing stale feature flags. This system integrates live experimentation data, engineer approval, isolated Git worktrees, and automated validation to ensure accurate and efficient cleanup.
Efficiency and Cost Savings
In an evaluation, the LLM system successfully generated usable pull requests for 45 out of 50 stale flags. Each cleanup averaged 13.8 minutes and cost $4.79, a substantial improvement compared to DoorDash's estimated manual cleanup time of one to two hours per flag.
The Challenge of Stale Flags
DoorDash's experimentation platform manages over 60,000 feature flags across approximately 623 repositories, with about 2,300 new flags created monthly. The company identified more than 1,000 stale flags, defined as flags unchanged for 90 days, still referenced in code, and not archived or excluded. The complexity of DoorDash's dependency-injected wrappers means a simple Boolean flag can require changes across five to 20 files, including tests.
LLM-Based Approach vs. Traditional Methods
Existing solutions, such as Uber's Piranha, use abstract syntax tree-based transformations. However, DoorDash found these approaches insufficient for its dependency injection patterns, where relationships are semantic rather than syntactically direct. The LLM-based system addresses this by understanding the semantic connections within the codebase.
Workflow and Agent Operations
The DoorDash workflow, built using Google's Agent Development Kit, operates in two phases. An orchestrator agent, running Claude Sonnet, retrieves stale flag tickets from Jira and gathers metadata. After engineer review, Claude Opus cleanup agents operate in isolated Git worktrees, with up to four agents running concurrently per repository. These agents locate flag references, determine cleanup strategies, modify code and tests, and run builds and coverage checks.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Reporting from
DoorDash developed a multi-agent large language model (LLM) system to automate the cleanup of stale feature flags across its codebase. This system significantly reduces the time and cost associated with manually removing unused flags, addressing a common challenge in large-scale software development.