Xcode moves into agentic coding with deeper OpenAI and Anthropic integrations

Apple is bringing agentic coding capabilities directly into Xcode. On Tuesday, the company announced the release of Xcode 26.3, which enables developers to use agentic coding tools — including Anthropic’s Claude Agent and OpenAI’s Codex — natively within Apple’s official app development environment.

The Xcode 26.3 Release Candidate is available today to all Apple developers through the developer website, with a broader rollout via the App Store planned for a later date.

The update builds on last year’s Xcode 26 release, which introduced support for ChatGPT and Claude inside Apple’s integrated development environment used by developers creating apps for iPhone, iPad, Mac, Apple Watch, and other Apple platforms.

With the addition of agentic coding, AI models can now leverage a broader set of Xcode features, enabling them to perform more advanced tasks and manage increasingly complex automation workflows.

These AI agents also gain access to Apple’s up-to-date developer documentation, ensuring that any code they generate relies on the latest APIs and follows Apple’s recommended best practices.

At launch, the agentic tools can help developers explore an existing project, understand its structure and metadata, build the project, run tests, identify errors, and fix issues where possible.

To support this release, Apple said it collaborated closely with both Anthropic and OpenAI to shape the overall experience. The company emphasised that significant effort went into optimising token usage and tool-calling behaviour to ensure agents operated efficiently within Xcode.

Xcode uses Model Context Protocol (MCP) to expose its capabilities to AI agents and connect them with its internal tools. As a result, Xcode can now work with any external agent that supports MCP for tasks such as project discovery, code changes, file management, previews and snippets, and accessing the latest documentation.

Developers who want to try agentic coding must first download the agents they plan to use through Xcode’s settings. They can then link their accounts with supported AI providers by signing in or adding an API key. A drop-down menu in Xcode lets developers select the model variant to use, such as GPT-5.2-Codex or GPT-5.1 mini.

Using a prompt box on the left side of the interface, developers can describe what they want to build or specify changes to their code in natural language. For example, they can instruct Xcode to add a new feature that relies on one of Apple’s frameworks and define how that feature should look and behave.

As the agent processes the request, it breaks it into smaller steps, making it easy for developers to follow what’s happening and see how the code evolves. The agent also searches for relevant documentation before writing code. All modifications are visually highlighted within the editor, and a project transcript panel shows what the agent is doing behind the scenes.

Apple believes this level of transparency could be invaluable for developers who are still learning how to code. To support that audience, the company is hosting a live “code-along” workshop on Thursday via its developer website, where participants can follow along in real time using their own copy of Xcode while learning how to work with agentic coding tools.

Once the agent completes its work, it verifies that the generated code functions as expected. Based on the test results, the agent can continue iterating on the project to resolve errors or address other issues. Apple noted that prompting the agent to think through its plan before writing code can sometimes improve outcomes by encouraging more deliberate planning.

If developers are unhappy with the results at any stage, they can easily roll back changes. Xcode automatically creates milestones whenever the agent modifies the project, allowing developers to revert to an earlier state as needed.