Agentic coding comes to Apple’s Xcode with agents from Anthropic and OpenAI

Apple has officially introduced agentic coding capabilities with the launch of Xcode 26.3. Announced on Tuesday, this update enables developers to utilize advanced coding tools, including Anthropic’s Claude Agent and OpenAI’s Codex, directly within Apple’s app development environment. Available now for all Apple Developers via the developer website, the Xcode 26.3 Release Candidate will soon be accessible through the App Store. This enhancement builds on the foundational integration of AI tools like ChatGPT and Claude in last year's Xcode 26 release, further enriching the development experience for applications across iPhone, iPad, Mac, Apple Watch, and other Apple devices. With the new agentic coding features, AI models can now leverage a broader range of Xcode functionalities to execute tasks and automate complex processes. These models are granted access to Apple’s developer documentation, ensuring they utilize the latest APIs while adhering to best practices throughout the development cycle. Initially, the agents will assist developers in exploring project structures, understanding metadata, building projects, and running tests to identify and rectify errors. To facilitate this launch, Apple collaborated closely with both Anthropic and OpenAI, focusing on optimizing token usage and tool interactions for seamless performance in Xcode. The integration employs the Model Context Protocol (MCP), which allows external MCP-compatible agents to engage with Xcode’s tools for various tasks, including project discovery and file management. Developers interested in the new coding features can download the desired agents from Xcode’s settings and connect their accounts by signing in or adding their API keys. A convenient drop-down menu within the app allows the selection of different model versions, such as GPT-5.2-codex and GPT-5.1-mini. Using natural language commands, developers can instruct the agent on the types of projects to build or modifications to implement in the code. As agents work, they decompose tasks into smaller, manageable steps, providing real-time visibility into the coding process and how the code evolves. The agents also seek necessary documentation before commencing coding. Visual highlights within the code showcase the changes, while a project transcript on the side offers insight into the underlying processes, particularly beneficial for novice developers. To support users in harnessing these agentic coding tools, Apple will host a “code-along” workshop on Thursday via its developer site, enabling participants to learn in real time alongside their own version of Xcode. At the conclusion of the coding process, the AI agent validates the functionality of the created code. If needed, it can further refine the project based on test results. Apple emphasizes that encouraging the agent to plan before writing code can enhance the overall coding experience. Additionally, developers retain the flexibility to revert to previous code versions at any point, as Xcode automatically creates milestones with each change the agent makes.