GitHub Copilot coding agent: an agent that works inside GitHub Actions

2025-05-19 · Publisher : GitHub · Domain : Cloud & DevOps

On 19 May 2025, GitHub unveils the Copilot coding agent: you assign it an issue, it works in a GitHub Actions environment and opens a draft pull request. CI workflows require human approval.

### Background On 19 May 2025, GitHub introduces the Copilot coding agent, an asynchronous agent built into the platform. The developer assigns it a GitHub issue; the agent works in an ephemeral environment powered by GitHub Actions, analyses the code, commits its changes to a draft pull request and exposes a session log. It is available to Copilot Enterprise and Copilot Pro+ subscribers. ### Key facts - From 4 June 2025, each model request made by the agent consumes one "premium request". - The agent can only push to branches it creates; the person who requested the work cannot approve its pull request. - GitHub Actions workflows do not run without a human's approval, leaving a check before CI. - Internet access is limited to a list of approved destinations. - The agent supports the Model Context Protocol (MCP) to reach external data sources and tools. ### Why it matters This is not only an assisted-coding topic: the agent runs inside CI, with its tokens, permissions and network, hence in DevOps territory. Teams must decide which permissions to give this new actor in the pipeline, how to frame its network access and who validates CI execution. GitHub's safeguards show the expected model: isolated branches, separation of roles and human approval. Billing per "premium request" from 4 June 2025 also makes the usage cost of an agent visible, which teams need to include in their FinOps tracking. ### Takeaway The coding agent becomes a CI actor: its place in pipeline permissions and reviews has to be defined.

Tags : Agents, CI/CD, MCP, DevOps, IA