Local AI Needs to be the Norm .

Cloud-hosted AI dependencies are making apps fragile, expensive, and privacy-invasive. Modern devices have dedicated neural engines capable of running local models for common tasks like summarization, classification, and extraction. Using Apple's FoundationModels framework as a concrete example, the post demonstrates how on-device AI can handle these tasks without sending user data to third-party servers. Apple's typed output pattern (using @Generable structs) is highlighted as a better engineering approach than parsing unstructured model responses. The core argument: local AI should be the default for tasks that transform user-owned data, with cloud models reserved only for cases where their additional capability is genuinely necessary.