Local AI Needs to Be the Norm
When developers add AI features to their apps, the default move is to call an OpenAI or Anthropic API. But this dependency is creating a more fragile world: apps stop working when the server crashes or a credit card expires, user data flows to third-party providers, and privacy policies become a substitute for actual trust.
A compelling blog post argues that local AI should be the default. Today’s mobile devices pack dedicated neural engines that are faster than data centers from a decade ago, yet most sit idle while apps wait for JSON responses from server farms in Virginia. Streaming user content to a third-party AI provider transforms a simple feature into an expensive distributed system, complete with data retention questions, consent audits, and compliance overhead.
The article showcases the Brutalist Report iOS client as a concrete example: it uses Apple’s built-in local model APIs to generate article summaries, with all processing happening on-device. No data leaves the user’s phone. No third-party accounts needed. No 30-day data retention clause. This not only simplifies the tech stack but eliminates the entire category of privacy promises and compliance burdens from the product.
Apple’s latest tooling also supports structured output — developers define Swift structs and the model returns typed data instead of unstructured text. This transforms AI from a novelty into a trustworthy subsystem within the app.
Local models can’t compete with cloud giants on general intelligence. But for the vast majority of real-world use cases — summarization, classification, extraction, rewriting, normalization — they’re already excellent. The principle is simple: use cloud models only when genuinely necessary. Otherwise, keep data where it belongs and build truly local-first applications.