"Should I add AI to my app?" Apple's Foundation Models framework has turned this simple question into a genuinely hard architecture decision. With on-device, Private Cloud Compute, and third-party models like Claude and Gemini now routable through a single API, the old default (just call a cloud API) is gone, and every AI-powered feature in your app deserves its own answer.

This talk gives you a practical framework for making that call feature by feature. We'll walk through four axes that actually matter in production: latency and offline behavior, capability ceiling (the "free 80%" vs. the "hard 20%"), privacy and user trust, and cost at scale. Then we'll apply the framework to a real app with multiple AI features, showing how the same decision tree produces different answers for smart replies, document summarization, and visual search, often within the same codebase.

You'll leave with a mental model for scoping AI features with stakeholders, a way to avoid over-committing to one model too early, and a clear-eyed look at what's actually new in iOS 27 versus what's still marketing. This is a strategy talk, not a syntax tour, ideal for app developers and tech leads who are about to have this exact conversation with their team.