GitHub Copilot Adds Claude and Codex
GitHub Copilot has integrated Anthropic's Claude and OpenAI's Codex models for all paid subscribers at no extra cost. This upgrade turns Copilot into a multi-model platform, letting engineers toggle between LLMs for code generation and debugging directly in their IDE. The move sets a new standard for developer tools, de-risking vendor lock-in and accelerating prototyping.
This integration marks a significant evolution from Copilot's origins. First launched in June 2021, the tool was initially powered exclusively by OpenAI Codex, a model trained on billions of lines of public code to turn natural language into code snippets. Over time, Copilot upgraded its base model to newer versions like GPT-4, continuously improving its code suggestion capabilities. The addition of Claude and Codex as distinct "agents" transforms Copilot from a single-model assistant into a platform. Developers can now run agents in parallel within the same IDE or on GitHub.com to compare different approaches to a problem before committing to a line of code. These agents share context from repository code, issues, and pull requests, eliminating the need to re-explain tasks. Anthropic's Claude excels at high-level reasoning and understanding entire codebases, thanks to its extremely large context window. This makes it ideal for strategic tasks like large-scale refactoring or comprehending legacy systems. In contrast, the latest iteration of OpenAI's Codex functions more like an autonomous software engineer, capable of taking on and completing entire development tasks independently. This multi-agent approach signals a shift in the engineering workflow from focusing on syntax to strategy and review. The developer's role is elevated to that of an architect, delegating implementation tasks to specialized AI agents. This changes the core work from writing boilerplate code to defining problems, choosing the right AI tool for the job, and validating the results. For engineers at startups, this model accelerates prototyping by allowing for rapid, parallel experimentation. Instead of a single developer being locked into one implementation path, they can assign the same issue to both Claude and Codex, then review and merge the best aspects of each AI-generated pull request. Access to these new agents is bundled into existing paid Copilot plans, including Pro and Business tiers, after an administrator enables them in the organization's settings. The integration works directly within VS Code, GitHub Mobile, and the GitHub web interface, centralizing AI collaboration within the existing developer workflow.
