Copilot's coding agent turns the issue queue into the prompt
GitHub's Copilot coding agent — generally available to paid subscribers — starts from an assigned issue, opens a draft pull request, works in its own environment through Actions, runs CI, and requests review. The code generation is not the interesting part. The entry point is.
Every competing agent starts from a prompt in an editor or a terminal, somewhere a developer has already decided to engage. Starting from an assigned issue means the work begins inside the tracker where prioritisation already happens. The agent inherits an existing workflow instead of asking for a new one, and that is worth more than several benchmark points.
It is the same structural move as Salesforce rebuilding Slackbot into an agent, and it is worth naming as a pattern: in agentic tooling, owning the queue where work is already tracked beats having the better model. Distribution is the moat, and incumbents own distribution.
Opening as a draft and running CI before requesting review is the detail that makes it usable rather than annoying. An agent that opens finished-looking pull requests without verifying them shifts work onto reviewers and gets switched off within a week. Gating on a green build is the difference between automation and noise.
The GitHub Blog — GitHub Copilot: the agent awakens → · GitHub Docs — Starting GitHub Copilot coding agent sessions → · CSW Solutions — GitHub Copilot coding agents: from suggest code to open the pull request →