Apple has introduced agentic coding to its flagship development tool, Xcode, with the announcement of Xcode 26.3. The update allows developers to use autonomous AI agents directly within Apple’s official app development environment. These agents include Anthropic’s Claude Agent and OpenAI’s Codex, which can operate inside Xcode to help developers build and test software.
The release candidate of Xcode 26.3 is now available to members of the Apple Developer Program through Apple’s developer website and is expected to be released on the App Store at a later date.
This update follows the earlier Xcode 26 release last year, which introduced support for tools such as ChatGPT and Claude within Apple’s integrated development environment. Xcode is used to create applications for iPhone, iPad, Mac, Apple Watch, and other Apple hardware platforms. With version 26.3, Apple is moving beyond basic code assistance and enabling AI tools to take more autonomous actions throughout the coding workflow.
Through agentic coding, AI models gain deeper access to Xcode’s features and can carry out more advanced tasks. These include using Apple’s latest developer documentation to identify current APIs and best practices, exploring project structure, understanding metadata, building applications, running tests, identifying errors, and fixing issues when they arise.
To support these capabilities, Apple said it worked closely with both Anthropic and OpenAI to design the integration. According to the company, significant effort went into optimising token usage and tool calling so that the agents operate efficiently within Xcode.
Xcode relies on the Model Context Protocol, or MCP, to expose its internal tools and capabilities to AI agents. This allows the built-in Claude Agent and OpenAI’s Codex to interact with project files, previews, documentation, and workflows. The protocol also enables compatibility with other external agents that support MCP, expanding flexibility for developers.
Developers who want to try the agentic coding features can download the agents they wish to use directly from Xcode’s settings. They can connect their OpenAI or Anthropic accounts by signing in or entering an API key, and then select their preferred model version from a drop-down menu within the app, such as GPT-5.2-Codex or GPT-5.1 mini.
A prompt box on the left side of the interface allows developers to describe what they want to build or what changes they want to make using natural language. For example, they can ask the agent to add a new feature using one of Apple’s frameworks and specify how it should appear and function. As the agent works, it breaks tasks into smaller steps so developers can follow what is happening and see how the code is evolving. This includes reviewing documentation before coding, visually highlighting changes in the code, and displaying a transcript that explains the agent’s actions.
Apple believes this level of transparency will be especially helpful for developers who are new to coding, as it makes both the reasoning and execution behind changes easier to understand. To support adoption, Apple plans to host a code-along workshop on its developer website, where participants can watch demonstrations and practise using agentic coding tools in real time with their own copy of Xcode.
After completing its work, the AI agent verifies that the code runs as intended. Based on the test results, it can continue refining the project to resolve errors or other issues. Apple noted that asking the agent to outline its plan before writing code can sometimes improve results by encouraging structured thinking in advance.
If developers are unhappy with the outcome, Xcode allows them to revert to earlier versions of their code at any time. The system automatically creates milestones whenever an agent makes changes, making it easy to roll back to a previous state.
