Faire uses swarm-coding with multiple GitHub Copilot background agents to accelerate large-scale engineering workflows
The problem
Faire engineers spent countless hours on tedious but essential maintenance tasks like cleaning up expired feature flags and migrating test infrastructure, with no scalable way to parallelize or automate this work.
Workflow diagram · grounded in source
1
Issue assigned to Copilot
Trigger
GitHub Copilot's background agent is triggered when a GitHub issue is assigned to it.
▾ source quote
“GitHub Copilot's background agent is an autonomous workflow that will generate a pull request when you assign it to a GitHub issue”
2
Expired setting finder
Ai action
An ExpiredSettingFinderAgent finds expired settings to clean up and manages allocation of cleanup tasks.
▾ source quote
“ExpiredSettingFinderAgent: An agent that can find expired settings to clean up, and manage allocation of the cleanup tasks”
3
Cleanup readiness assessment
Ai action
A SettingCleanupReadinessAgent determines the stage of cleanup and generates instructions for the next step.
▾ source quote
“SettingCleanupReadinessAgent: An agent that determines the stage of clean up, and generates instructions for the next step”
4
Background agent executes changes
Ai action
GitHub Copilot autonomously executes code changes and incrementally updates a placeholder pull request with a TODO list and progress.
▾ source quote
“a specialized GitHub action to run in the background, which will put up a placeholder pull request that it incrementally updates with a TODO list and the progress towards it”
5
Build failure feedback to agent
Validation
Build failures are automatically summarized and surfaced to Copilot so it can iterate.
▾ source quote
“Summarize the build failures for Copilot — automatically surfaces problems to Copilot to iterate”
6
Two-human review required
Human review
Two human reviews are required on Copilot-authored pull requests — the assignee's initial review plus another human.
▾ source quote
“Require two reviews on PRs authored by Copilot — the assignee's initial review, plus another human”
7
Time saved reported via Slack
Feedback loop
When a PR is merged, the assignee is pinged on Slack to report how much time the Copilot PR saved them.
▾ source quote
“When a PR is merged, we ping the assignee on Slack to ask how much time it saved them”
Reported outcome
Within just over a month, 18% of the engineering team adopted GitHub Copilot, over 500 Copilot pull requests were merged, Copilot users saw a 25% increase in PR volume, and the average reported time saved was 39.6 minutes per PR.
Tell us your scale, team, and constraints. We'll show what changes at your size, what fails at your scale, and whether this case is a fit, needs adaptation, or won't scale to you. Free demo, no signup.
Within just over a month, 18% of the engineering team adopted GitHub Copilot, over 500 Copilot pull requests were merged, Copilot users saw a 25% increase in PR volume, and the average reported time saved was 39.6 min…
Engineering team Copilot adoption: 18%; Copilot pull requests merged: over 500; PR volume increase for Copilot users: 25%; Average time saved per PR: 39.6 minutes (source-reported, not independently verified).
How is this back office ops AI workflow structured?
Issue assigned to Copilot → Expired setting finder → Cleanup readiness assessment → Background agent executes changes → Build failure feedback to agent → Two-human review required → Time saved reported via Slack.
This case is one data point. Whether its pattern fits you depends on your volumes, your stack, and your exception load — that comparison is the step no case study can do for you.