GitHub Expands Copilot With Claude and Codex for Business and Pro Users

GitHub adds model choice to Copilot workflows

GitHub has made Claude and Codex available to Copilot Business and Copilot Pro users, expanding the set of AI coding agents that can be used within GitHub development workflows.

In its February 26 changelog announcement, GitHub said Claude, Codex and Copilot can run asynchronous sessions that work on development tasks and produce draft pull requests. The company also highlighted enterprise controls, policy management and audit logging for organizations using the tools.

The update places several prominent coding models in a workflow centered on issues, repositories and pull-request review. Rather than limiting AI assistance to inline suggestions in an editor, the approach allows a developer to assign a larger task, inspect the proposed changes and decide whether to merge them.

Coding tools are moving toward multi-agent work

GitHub’s announcement reflects a broader move among developer tools toward environments where teams can work with multiple agents that have different capabilities and levels of access.

Visual Studio Code described that model in a February 5 post on multi-agent development. According to Microsoft’s editor team, VS Code can run Claude, Codex and Copilot as local, background or cloud agents. The post also described parallel subagents, MCP Apps and specialized agents configured with distinct permissions.

Access controls matter because an agent assigned to a software task may need to read source code, edit files, run development tools or connect to external services. Giving different agents different permissions can help teams define what each system is allowed to access or change.

GitHub’s focus on policies and audit logging addresses the same operational concern. For business users, adopting coding agents involves more than choosing a model: organizations also need visibility into how the tools are used and controls that fit their repository and review practices.

Delegation, review and administration

Other vendors have emphasized similar features. Coder, in its announcement for Coder Agents, said its self-hosted product provides centralized administration for models, prompts, MCPs and skills, while using network-isolated workspaces. Coder also said agents can be triggered through APIs and connected with CI/CD systems and GitHub Actions.

WIRED reported that Cursor 3 lets developers launch agents for coding work, including cloud-based feature development that can later be reviewed locally. The publication positioned Cursor’s release amid competition with Anthropic’s Claude Code and OpenAI’s Codex.

These products share an emphasis on delegated work: an AI system takes on a defined engineering task, while a developer remains responsible for reviewing the output. GitHub’s addition of Claude and Codex gives Copilot Business and Pro customers more choice over which agent participates in that process.

What the change means for teams

For individual developers, having multiple models available may support different preferences for task complexity, coding style or workflow. For organizations, the practical question is whether asynchronous agent sessions, draft pull requests and governance controls can be incorporated into existing permission structures and code-review procedures.

GitHub’s announcement does not suggest that AI-generated changes should bypass review. Draft pull requests remain proposals that developers can inspect, revise and approve. The change instead broadens the range of systems that can help generate those proposals inside GitHub’s development environment.