DoorDash agent cleanup and Copilot engagement metrics land the same week
InfoQ said DoorDash built a multi-agent LLM workflow to scan more than 60,000 stale feature flags across 623 repositories, pulling live experiment data through MCP without reporting a cleanup yield. GitHub added 28-day Copilot feature engagement counts to enterprise impact dashboards the day before.
Artificial Intelligence··Night
Sixty thousand flags, one MCP pipe
InfoQ's 18 September report said DoorDash automated stale feature-flag cleanup with a multi-agent LLM system spanning more than 60,000 flags and 623 repositories. The workflow pulls live experiment data through MCP rather than relying on static lists alone. InfoQ timestamps the write-up at 12:00 UTC and notes the announcement as a response to growing software-agent sprawl. The summary does not state a cleanup yield, an error rate, or how often human reviewers veto an agent decision.[1]
GitHub counts who keeps using Copilot features
GitHub's 17 September changelog added copilot_feature_engagement to impact dashboards and to enterprise and organization 28-day aggregate reports. The field counts Copilot features that active users touched on at least two days within the rolling 28-day window, and totals_by_feature splits code completion, agent edit, passive and active review, cloud agents, CLI and the Copilot app. A user can appear under more than one feature bucket. The counter measures repeated use, not whether an agent task completed safely or a flag deletion was correct.[2]
Two dashboards, different questions
InfoQ documents an operational agent workflow at DoorDash; GitHub documents adoption telemetry for Copilot features inside customer repositories. Both reports arrived within a day and both speak to software agents at scale, yet neither ties engagement counts to cleanup outcomes or safety review load. InfoQ leaves yield blank; GitHub leaves correctness outside the metric schema. The pair shows instrumentation moving faster than shared definitions of success for agent work.[1], [2]
