Xcode Intelligence

Xcode 26 was introduced to the world as part of WWDC25, in June 2025. At the time of writing it is available as a developer beta, but will be released in Autumn. One of the major new features in Xcode is the introduction of Apple Intelligence, in the form of integration with a range of different LLMs.

Since this isn’t fully available at the time of writing, we’ll briefly discuss the features promised, and how they relate to what you’ve already learnt as part of this module.

LLM Integration

The backbone to the new Xcode Intelligence integration is, as with all the other tools we’ve seen, a Large Language Model (LLM). The primary method of interaction is via a new chat pane—a concept familiar from our look at Cursor.

By default, Xcode integrates with ChatGPT, and has a limited number of free requests per day. If you have a ChatGPT subscription, then you’re able to login and get much greater access.

You can also link to a whole host of other LLM providers including Anthropic, as well as linking to locally-running LLMs such as Ollama. In this sense it is very similar to Cursor or Copilot in VSCode.

You can then use the chat window to interact with your code in the same way as we’ve seen repeatedly across this course: you can ask questions about your code, or generically about writing Swift. You can choose whether Xcode should send project context it thinks relevant, manually choose context to send, or to not send any context at all. The interface for managing context appears, at first glance, to be more intuitive than seen in Cursor, but it’s difficult to know how easy it is to use at this stage.

One of the ways you can provide context is through the “Coding Tools” interface, where you can interact directly with the LLM within the code file, rather than using the chat window. This offers commonly used features like “document this code”, or an inline chat to ask questions about specific code in a specific file.

Applying Changes

Again, similar to the tools we’ve already reviewed, you can choose to not apply code changes (chat mode), to selectively review and apply them, or to apply them all in a big batch.

When applying changes, the AI-generated changes are clearly highlighted for review, and Xcode records the history of changes it has applied, so you can easily review and undo at a later stage.

Agentic Coding

The thing that appears to be missing at this stage is some kind of agent-mode for writing code. We saw how powerful this can be in Cursor and Claude Code, but it is currently absent from the Xcode 26 beta. This could well be a conscious decision, which would feel quite Apple-like—why does the user care how the code was generated, just that it has been and that it is good. It is quite possible that agent mode will be added, but without the user having to know anything about it.

Conclusion

It’s really nice to see that many of the techniques and approaches that we’ve explored in this module will be integrated into Xcode 26. It’s quite possible that this might be just enough for most developers to access these tools, but in a world changing as fast as AI-assisted coding, the latest and greatest developments will likely always start outside of Xcode.

Once Xcode 26 becomes generally available, it’ll be a good time to revisit this, with some real-world examples, and to put Xcode Intelligence through its paces.