GitHub Copilot Coding Agent launched at Microsoft Build in May 2025 as a distinct product from GitHub Copilot's autocomplete and chat features. Where Copilot assists developers as they write code, the Coding Agent operates independently: a developer assigns a GitHub issue to Copilot, the agent researches the repository, plans the implementation, creates a branch, writes the code, runs the test suite via GitHub Actions, and opens a draft PR — all in the background while the developer works on other tasks.
Key capabilities
Issue-to-PR pipeline — The complete workflow starts with assigning a GitHub issue (via web, mobile, VS Code, JetBrains, Slack, Linear, Jira, MCP, or REST API) and ends with a reviewable draft PR. The agent handles planning, implementation, and test execution autonomously.
GitHub Actions integration — The agent runs tests, linters, and CI checks via GitHub Actions in its own Actions runner (60-minute session limit), iterating on failures until tests pass or the time limit is reached.
Multi-source assignment — Issues can be assigned to the Coding Agent from GitHub Issues, Slack messages, Linear tickets, Jira issues, or via MCP or REST API, integrating with existing team workflows.
Security separation — The person who assigns an issue cannot be the same person who approves the resulting PR, enforcing a four-eyes review principle on all AI-generated code.
Autonomy level
Level 4 — Near-autonomous. The agent executes the full implementation cycle independently. Human code review and PR approval is required before any changes merge; the agent operates within GitHub's existing pull request governance model.
Strengths
- Native GitHub integration — no additional tooling or setup required
- Available to all paid GitHub Copilot plans (Pro, Business, Enterprise)
- Broad assignment channels: GitHub, VS Code, JetBrains, Slack, Linear, Jira, MCP
- GitHub Actions test runner provides real CI validation before human review
- Security separation between assigner and reviewer built in
Limitations
- Web/cloud only — no local execution option
- 60-minute session limit per task
- Requires GitHub Copilot subscription (not available on Free plan)
- Usage-based billing (AI Credits) introduced June 2026 — complex cost modelling
- Not open source; no visibility into agent execution internals