Apple Shares 'Inside Apple Intelligence and Xcode' Special Presentation [Video]

Apple has released a special presentation from WWDC26 that takes developers inside the architecture of Apple Intelligence and the new tools built into Xcode. Filmed at the Steve Jobs Theater, the detailed video breaks down Apple's latest AI frameworks, Siri integrations, and agentic development tools.



A major portion of the presentation focuses on the agentic coding capabilities inside Xcode 27. Apple positioned the new assistant as an active collaborator rather than a simple autocomplete tool. It features a dedicated planning mode where developers can discuss an implementation before any code is written. The agent can then apply changes across multiple files, generate plans and previews alongside the conversation, and write and run tests to verify its work. Support for plugins also lets developers connect Xcode to external models, tools, and services.


The video also explores how applications can interact directly with Siri using the App Intents framework. By adopting specific entities and schemas, developers can expose their app's content and actions to Apple Intelligence. This allows Siri to search app content, perform actions through natural language requests, and incorporate app information into more contextual experiences. The session highlights new entity annotations that enable onscreen awareness, giving Siri the ability to understand what a user is looking at and suggest relevant actions.

For teams looking to deploy generative features, the Foundation Models framework provides a unified native Swift API. Developers can use Apple's on-device models, the new Private Cloud Compute server model, or third-party language models through the same framework. Apple also highlighted new Foundation Models capabilities including image input, support for multi-model agentic experiences with shared context, integration with third-party models such as Gemini and Claude through Swift packages, and plans to open-source the framework later this summer.

When developers need to run their own custom models locally, Apple is pointing them to the new Core AI framework. Built to take advantage of Apple Silicon, Core AI enables AI workloads to run across the CPU, GPU, and Neural Engine through a modern Swift API. To help engineers troubleshoot local deployments, Apple introduced a Core AI Debugger application that visualizes computation graphs, inspects tensor values, and traces data from the Core AI model back to the original Python source code.

Apple also introduced a new Evaluations Framework integrated directly into Xcode, giving teams a structured way to automatically test model outputs against custom grading rubrics and quality metrics. The framework can use language models as judges, synthesize test samples, aggregate results across large data sets, and help developers refine prompts through what Apple calls evaluation-driven development.


The presentation wraps up by highlighting updates to the MLX framework, Apple's open-source machine learning library. The spotlight here is on MLX Distributed, a tool that allows researchers to run massive language models that require more memory than a single machine can hold. By connecting multiple Macs into a local compute cluster, developers can scale training and inference workloads beyond the limits of a single machine.