Apple is introducing agentic coding capabilities directly within Xcode. This Tuesday, the company unveiled Xcode 26.3, enabling developers to utilize agentic tools such as Anthropic’s Claude Agent and OpenAI’s Codex from within Apple’s official app development suite. The Xcode 26.3 Release Candidate is now accessible to all Apple Developers via the developer website, with an App Store release to follow shortly.

This update builds upon the Xcode 26 release from last year, which initially integrated support for ChatGPT and Claude into Apple’s integrated development environment (IDE). This IDE is used for creating applications across Apple's ecosystem, including iPhone, iPad, Mac, and Apple Watch. The new agentic integration allows AI models to access a broader range of Xcode's functionalities, enabling them to handle more sophisticated automation tasks. These models will also be able to reference Apple's current developer documentation, ensuring they employ the latest APIs and adhere to recommended best practices during development.

At launch, these AI agents can assist developers by exploring a project's structure and metadata, building the project, and executing tests to identify and resolve any errors.

In preparation for this launch, Apple collaborated extensively with both Anthropic and OpenAI to craft the new experience. The company emphasized significant efforts to optimize token usage and tool calling, ensuring the agents operate efficiently within Xcode. The system utilizes MCP (Model Context Protocol) to expose Xcode's capabilities to the agents and connect them with its various tools. Consequently, Xcode can now interface with any external MCP-compatible agent for tasks like project discovery, file management, previews, snippets, and accessing up-to-date documentation.

Developers interested in testing the agentic coding feature should first download their preferred agents through Xcode's settings. They can also link their accounts with AI providers by signing in or adding an API key. A drop-down menu within the application lets developers select their desired model version, such as GPT-5.2-codex versus GPT-5.1-mini. Using a prompt box on the left side of the screen, developers can instruct the agent in natural language, describing the project they wish to build or the specific code changes they want to implement. For example, a user could direct Xcode to integrate a feature using one of Apple's frameworks, specifying its appearance and functionality.

As the agent begins its work, it decomposes tasks into smaller, manageable steps, making it easy to track progress and code modifications. It proactively searches for necessary documentation before starting to code. Changes are visually highlighted within the code editor, and a project transcript on the side of the screen provides insight into the underlying processes. Apple believes this transparency will be especially beneficial for new developers learning to code.

To support this, the company is hosting a "code-along" workshop this Thursday on its developer site, where participants can watch and learn how to use agentic coding tools in real-time alongside their own copy of Xcode. Upon completing its tasks, the AI agent verifies that the generated code functions as intended. Based on test results, the agent can further iterate on the project to correct errors or address other issues. Apple noted that prompting the agent to think through its plans before writing code can sometimes enhance the process by encouraging pre-planning.

Furthermore, if developers are dissatisfied with the results, they can easily revert their code to its original state at any time, as Xcode automatically creates milestones each time the agent makes a change.