Research, plan, and code with Copilot cloud agent

GitHub on April 1, 2026 expanded Copilot cloud agent beyond its earlier pull-request-centric workflow. The company said the service, formerly known as Copilot coding agent, can now work on a branch without opening a pull request first, giving developers more control over when to move from experimentation to review.

Branch-first instead of PR-first

Before this change, using Copilot cloud agent meant opening a pull request up front. GitHub now allows the agent to generate code on a branch without immediately creating one. Developers can inspect the full diff, iterate on the result, and only create a pull request when they decide the work is ready for review. GitHub also says users who already know they want a pull request can ask for that in the prompt and have Copilot open one when the session completes.

Planning before code

A second change is implementation planning. GitHub says users can ask Copilot to generate a plan before any code is written, review the proposed approach, and either approve it or send feedback. Once the plan is approved, the agent uses that plan to guide the implementation. This is a notable workflow adjustment because it makes design review a first-class step rather than something that happens after code is already produced.

Repository research as a separate task

The release also introduces deep research sessions. GitHub says developers can ask broad questions about their codebase and get answers grounded in repository context. The feature is meant for investigation rather than direct code generation, and GitHub says the same research flow can also be started from Copilot Chat conversations.

Availability

GitHub says the new functionality is available through agent entry points such as the Agents tab in the repository and Copilot Chat. Copilot cloud agent is included in all paid Copilot plans, while Copilot Business and Copilot Enterprise customers need an administrator to enable the feature. Taken together, the release moves Copilot closer to a full development workflow that covers investigation, planning, implementation, and review instead of treating code generation as a single isolated step.