GitHub Copilot Cloud Agent gets private sessions and model choice
GitHub's Copilot Cloud Agent (recently renamed from Copilot Coding Agent) now supports private pre-PR sessions, model selection across OpenAI and Anthropic models, and natural-language merge conflict resolution via `@Copilot`.
Score breakdown
The shift to private pre-PR sessions and on-demand `@Copilot` commands in PRs gives developers more control over when and how the agent's work becomes visible to their team, reducing friction in agentic coding workflows.
- 01Copilot Cloud Agent was renamed from Copilot Coding Agent as of last week before the video's publication.
- 02The agent no longer automatically opens a PR when a task starts — users now work in a private session first.
- 03The agents tab on GitHub serves as a central hub for submitting tasks and tracking agent activity.
GitHub's Copilot Cloud Agent — renamed from Copilot Coding Agent as of last week — has received a set of notable updates showcased in a GitHub Checkout video featuring Tim Rogers. The most significant behavioral change is the removal of automatic PR creation: previously, clicking "start task" would immediately open a pull request, but the agent now works in a private session first, letting developers review and iterate before deciding to open a PR. This also enables new use cases beyond code writing, such as asking the agent to plan before implementing, or to perform deep research on a codebase and return answers in logs rather than a PR.
The agents tab on GitHub.com acts as the central hub for interacting with agents, supporting task submission and activity tracking.
The agents tab on GitHub.com acts as the central hub for interacting with agents, supporting task submission and activity tracking. When using the Copilot agent, users can choose from multiple models from both OpenAI and Anthropic via a model picker. Claude and Codex agents are also available as alternatives. The video demonstrates submitting a task — implementing rate limiting at 60 requests per IP per hour — using Claude Opus, and shows how `@Copilot` can be mentioned directly in a PR to handle merge conflicts on demand.
The video also previews upcoming capabilities including automatic CI/CD fixes and a public API, though details on those features are not fully disclosed in the available transcript.
Key facts
- 01Copilot Cloud Agent was renamed from Copilot Coding Agent as of last week before the video's publication.
- 02The agent no longer automatically opens a PR when a task starts — users now work in a private session first.
- 03The agents tab on GitHub serves as a central hub for submitting tasks and tracking agent activity.
- 04Users can select from multiple models from OpenAI and Anthropic via a model picker; Claude and Codex agents are also available.
- 05Typing `@Copilot fix the merge conflicts` in a PR triggers the agent to handle conflicts automatically.
- 06New use cases enabled by private sessions include pre-implementation planning and deep codebase research without requiring a PR.
- 07A sneak peek at upcoming features includes automatic CI/CD fixes and a public API.