How contextual understanding actually works
After the WWDC 2026 keynote, Craig Federighi participated in a press Q&A where he clarified the technical workings of Siri AI's contextual understanding. According to Tom's Guide and confirmed by TechRepublic, Federighi described Siri AI as "deeply integrated into your experience, understanding what's on screen" and emphasized that while interactions are conversational, the system is designed as "an extension of your system experience, deeply integrated into your flow."
The most technically relevant point is the distinction between on-device processing — which handles on-screen context understanding, cross-app linking, and session memory building — and the use of Private Cloud Compute for requests that exceed local model capabilities. This hybrid architecture is what allows Siri AI to read an email, understand a calendar event, and suggest a Maps action without user data ever leaving the private sphere in a non-anonymized form.
Federighi also confirmed that on-screen comprehension is distinct from Visual Intelligence: the former operates in the background across the entire system, while the latter is explicitly activated by the user via camera or screenshot.