Claude 和 Codex 助力 Copilot Business/Pro：增强型 AI 编码

AI 编码代理：Claude 和 Codex 拓展 GitHub Copilot 访问权限

GitHub 正在通过集成 Anthropic 的 Claude 和 OpenAI 的 Codex 等尖端模型，显著增强其 AI 驱动的开发者工具 Copilot 的功能。这一关键性的扩展，是为 Copilot Business 和 Copilot Pro 客户宣布的，此前已向 Enterprise 和 Pro+ 层级开放。现在，更广泛的开发者可以直接访问这些复杂的 AI 编码代理，从而彻底改变在熟悉的 GitHub 生态系统中编写、审查和调试代码的方式。

Claude 和 Codex 的加入，使 GitHub Copilot 成为一个更强大、更通用的助手，提供多样化的 AI 视角和专业优势，以应对复杂的编码挑战。此举突显了 GitHub 致力于将最先进的 AI 直接集成到开发者工作流中的承诺，预示着一个前所未有的效率和创新未来。

GitHub 工作流的无缝集成

这种集成的真正力量在于它在所有核心 GitHub 环境中的无缝存在。无论您是通过 github.com 上的网页界面协作项目，还是使用 GitHub Mobile 随时随地处理问题，抑或是在 VS Code 中深入编码，Claude、Codex 和 Copilot 都可作为统一的编码代理使用。这意味着开发者可以在不同平台之间切换而不会丢失上下文，因为这些代理会保持对您工作的共享历史记录和理解。

这种普遍可用的访问消除了障碍，使开发者能够在最需要 AI 辅助的时候和地点利用它。从生成复杂的代码片段到为棘手的 bug 提出解决方案，甚至起草拉取请求，这些代理都成为开发生命周期中不可或缺的一部分，加速生产力并培养更直观的编码体验。最重要的是，不需要额外的订阅；Claude 和 Codex 的访问权限已完全包含在您现有的 Copilot 订阅中，这为现有用户带来了显著的附加价值。

企业级 AI 的统一控制与治理

对于组织和个人开发者而言，集成多个强大的 AI 模型需要一个稳健的管理和安全框架。GitHub 通过为所有代理提供一个单一的共享平台来解决这个问题，确保统一的治理、共享的上下文和内存。这种架构允许 Claude、Codex 和 Copilot 以对项目整体的理解来运行，访问仓库代码和历史记录、问题、拉取请求、Copilot Memory 以及仓库特定的指令和策略。

这种受控环境的核心是 GitHub 的企业控制，其特点是通过现已普遍可用的 Agent Control Plane 实现集中启用、策略管理和全面的审计日志记录。这个复杂的系统确保企业可以自信地部署 AI 代理，在赋予开发团队高级 AI 功能的同时，保持监督并遵守内部合规标准。对于采用代理工作流的企业来说，这种控制级别至关重要。

分步指南：为您的团队启用 Claude 和 Codex

激活这些强大的 AI 编码代理是一个直接的过程，尽管 Copilot Pro 和 Copilot Business 账户之间略有不同，这反映了个人与组织控制的不同需求。

功能 / 账户类型	Copilot Pro	Copilot Business
启用流程	用户级别控制	企业和组织级别的管理员控制
步骤 (Pro)	1. 进入 Copilot 编码代理设置。 2. 选择仓库。 3. 将“Claude”、“Codex”或两者都打开。	-
步骤 (Business)	-	企业：前往“Enterprise AI Controls”→“Agents”。在“Partner Agents”下，启用 Claude/Codex。组织：前往“Settings”→“Copilot”→“Coding agent”。在“Partner Agents”下，启用 Claude/Codex。
权限	个人用户选择	由企业/组织管理员集中控制
额外费用	无（已包含在现有订阅中）	无（已包含在现有订阅中）

这种分层方法确保个人开发者和企业管理员都对 AI 代理的访问和使用拥有必要的精细控制，以适应其特定的运营需求和安全策略。

利用 AI 代理提升开发者生产力

一旦启用，Claude 和 Codex 将开启开发者生产力的新维度。与这些代理交互的方法直观且已集成到日常开发任务中。

启动会话：

GitHub（网页版和移动版）： 只需在已启用的仓库中打开“Agents”标签页，或使用主标题栏中的代理下拉菜单。在 GitHub Mobile 上，通过“Agents”视图或点击“+”图标进行访问。输入您的请求，选择您偏好的代理（Claude、Codex 或 Copilot），然后提交。会话将异步运行，提供实时进度和详细日志以供审查。
VS Code： （需要 VS Code 1.109 或更高版本）从标题栏或命令面板打开“Agent sessions”。您可以选择“Local”以获得快速、交互式帮助，“Cloud”用于在 GitHub 上运行自主任务，或“Background”用于异步本地任务（仅限 Copilot）。

将代理分配给问题和拉取请求： 开发人员现在可以将问题分配给特定的代理，例如 @copilot、@claude 或 @codex，甚至可以分配给多个代理以比较它们的处理方法。这些代理随后可以生成草稿拉取请求以供审查，从而简化开发的初始阶段。此外，在拉取请求评论中提及代理（例如，@claude 以寻求复杂逻辑的帮助）允许开发人员直接在其现有工作流中请求更新或改进。所有代理输出都以草稿、可审查工件的形式出现，确保开发人员对最终代码保持完全控制。这包括利用像 Claude Opus 4.6 这样的模型对特定任务的细致理解。

融合先进 AI 的协作编码未来

Claude 和 Codex 集成到 GitHub Copilot 中标志着 AI 辅助软件开发迈出了重要一步。通过在一个统一、受控的平台上提供多种强大的 AI 模型选择，GitHub 不仅仅是提供工具；它正在培育一个生态系统，让开发人员能够充分利用人工智能的潜力，更快、更高效地进行创新。

此次扩展使更广泛的开发者社区能够尝试不同的 AI 方法，比较结果，并最终为每个特定任务选择最佳的 AI 合作伙伴。随着 AI 模型的不断发展，在现有工作流中无缝集成和管理它们的能力将至关重要。GitHub 的战略确保其用户在这场革命中保持领先地位，配备最先进的 AI 编码代理，以应对现代软件工程的复杂性。编码的未来是协作的、智能的，并且日益由 AI 驱动。

原始来源

https://github.blog/changelog/2026-02-26-claude-and-codex-now-available-for-copilot-business-pro-users/

常见问题

What are Claude and Codex, and how do they enhance GitHub Copilot's capabilities?

Claude by Anthropic and OpenAI Codex are advanced AI coding agents now integrated into GitHub Copilot. They significantly enhance Copilot's capabilities by offering diverse AI models for code generation, review, and issue resolution. Developers can leverage the distinct strengths of Claude for nuanced understanding and robust reasoning, and Codex for its deep expertise in code generation, enabling a more versatile and powerful AI-assisted development experience. This expansion ensures that Copilot users have access to a broader spectrum of AI intelligence for various coding challenges, from complex problem-solving to efficient code completion.

Which GitHub Copilot subscription tiers now have access to Claude and Codex agents?

Following an earlier rollout to Copilot Enterprise and Pro+ customers, GitHub has now expanded access to Claude and OpenAI Codex to all Copilot Business and Copilot Pro subscribers. This means a much wider range of individual developers and business teams can now utilize these powerful AI coding agents directly within their existing GitHub workflows, without requiring any additional subscriptions. The goal is to democratize advanced AI assistance in coding, making high-quality, diverse AI models available to a larger user base for enhanced productivity.

How do these AI coding agents integrate into existing GitHub workflows and development environments?

Claude and Codex are designed for seamless integration across the entire GitHub ecosystem. They can be utilized directly on github.com (web), GitHub Mobile, and within VS Code. This deep integration allows developers to use the agents for reviewing code, triaging issues, generating solutions, and iterating on code – all within their familiar GitHub environments. The agents maintain shared history and context across these surfaces, ensuring a consistent and efficient experience whether you're at your desktop or on the go. This level of integration streamlines the developer workflow significantly.

What kind of data and context do Claude and Codex agents have access to within GitHub?

Operating on a unified platform within GitHub, these AI agents have secure and governed access to crucial contextual data. This includes access to repository code and its history, relevant issues, and pull requests. Furthermore, they can leverage Copilot Memory and adhere to repository-specific instructions and policies. This rich context allows Claude and Codex to provide highly relevant and accurate assistance, understanding the specific codebase and ongoing development tasks, while operating within the enterprise's defined controls and audit logging through the Agent Control Plane.

What administrative controls are available for managing Claude and Codex access for Copilot Business users?

For Copilot Business customers, robust administrative controls are in place through the generally available Agent Control Plane. To enable Claude and Codex, administrators must first activate them at the enterprise level via 'Enterprise AI Controls' → 'Agents' under 'Partner Agents'. Subsequently, organization-level settings must also be configured under 'Settings' → 'Copilot' → 'Coding agent,' where 'Partner Agents' can be enabled. These multi-layered controls ensure centralized enablement, policy management, and comprehensive audit logging, giving businesses granular oversight of AI agent usage.

How can developers initiate and manage sessions with Claude and Codex within GitHub and VS Code?

Developers can initiate sessions with these agents in multiple ways. On GitHub (web and mobile), they can open the 'Agents' tab in an enabled repository or use the agents dropdown in the main header. In VS Code (version 1.109 or later), 'Agent sessions' can be opened from the title bar or Command Palette, offering 'Local' for interactive help, 'Cloud' for autonomous tasks on GitHub, or 'Background' for asynchronous local tasks (Copilot only). Requests are entered, an agent is chosen, and sessions run asynchronously with live progress and detailed logs, ensuring transparency and control.

Are there any additional costs or subscriptions required to use Claude and Codex with GitHub Copilot?

No, there are no additional subscriptions or costs required for existing Copilot Business and Copilot Pro users to access Claude and Codex. GitHub has integrated these powerful AI coding agents directly into the existing Copilot subscription. Each coding agent session currently consumes one 'premium request' during its public preview phase. This ensures that users can immediately leverage the enhanced capabilities of these advanced AI models without incurring extra fees, maximizing the value of their current Copilot investment.

How does GitHub ensure the security and governance of sensitive code when using these external AI models?

GitHub places a strong emphasis on security and governance. Claude and Codex operate within GitHub’s enterprise controls, utilizing a unified platform with shared context and memory, but crucially, all activity is subject to centralized enablement, policy management, and audit logging via the generally available Agent Control Plane. This ensures that usage of these external AI models adheres to organizational policies and provides transparency and traceability. Agents access repository data only as permitted, and all output is presented as draft, reviewable artifacts, allowing developers to maintain control and oversight of their code.

保持更新

将最新AI新闻发送到您的收件箱。