This is GitHub’s announcement of GitHub Copilot coding agent, published in May 2025 and updated later the same month. Its central claim is an interface one: the way to delegate work to the agent is to assign a GitHub issue to Copilot, exactly as an issue would be assigned to a team member, after which the agent works in the background and delivers its output as a draft pull request.
The post’s other emphasis is that delegating to an agent should not require relaxing a repository’s existing controls. It sets out a default policy set — branch restrictions, a bar on the requester approving the agent’s own pull request, a limited internet allowlist, and CI workflows that do not run without human approval — and presents these as the reason an agent can be added to a team without weakening its security posture.
Key Points
- The agent is invoked by assigning one or more GitHub issues to Copilot, from github.com, GitHub Mobile or the GitHub CLI; it can also be asked to open a pull request from Copilot Chat on GitHub or in VS Code.
- Once assigned, it adds an eyes emoji reaction, then boots a virtual machine, clones the repository, configures the environment and analyses the codebase using retrieval augmented generation powered by GitHub code search.
- It pushes its changes to a draft pull request as git commits and updates the pull request description as it goes, with its reasoning and validation steps visible in session logs.
- GitHub states the agent excels at low-to-medium complexity tasks in well-tested codebases — adding features, fixing bugs, extending tests, refactoring, improving documentation — which is a scoping claim, not a measured result.
- Model Context Protocol servers can be configured in the repository’s settings to give the agent access to data and capabilities outside GitHub, including GitHub’s own MCP server.
- Vision model support lets the agent read images attached to the issues assigned to it, so a bug screenshot or a feature mockup can be part of the task.
- After finishing, it tags the requester for review and picks up review comments automatically, proposing code changes in response.
- GitHub’s stated reason for choosing GitHub Actions as the compute layer is scale: Actions, introduced in 2018, is described as the largest CI/CD ecosystem in the world, with more than 25,000 actions in the GitHub Marketplace and more than 40 million jobs run every weekday.
- Four default policies are applied: the agent may push only to branches it created; the developer who asked for the pull request cannot approve it, so existing required-review rules are honored; the agent’s internet access is limited to a customizable trusted list; and GitHub Actions workflows do not run without human approval.
- At announcement the agent was available to Copilot Enterprise and Copilot Pro+ customers, opted in per repository, with an additional organization policy for Enterprise. From June 4, 2025 it was to consume one premium request per model request.
Context
The post is GitHub’s own product announcement, so its characterizations of the agent’s strengths are the vendor’s. It carries two customer quotes supplied for the announcement — a DevEx lead at EY describing the agent as letting developers assign work that would otherwise detract from deeper work, and an engineering vice president at Carvana describing it as converting specifications to production code in minutes — which are relayed by GitHub rather than independently reported.
The agent is presented as the continuation of an internal effort the post calls Project Padawan, and as one point on a line GitHub draws through its own products: code completions, next edit suggestions, chat, agent mode, and now a background agent.