Local AI Should Be the Norm, Not Cloud Dependencies
The author argues that developers are overly reliant on cloud-hosted AI models (OpenAI, Anthropic) for app features, creating fragile, privacy-invasive software. They advocate for on-device AI processing, citing Apple's local model APIs as a viable alternative. A concrete example is given: the Brutalist Report iOS app generates article summaries entirely on-device using Apple's local models, avoiding server detours, data retention issues, and vendor dependencies. The author emphasizes that local AI excels at transforming user-owned data (summarization, classification, extraction) and that most features do not require the intelligence of cloud models. They highlight Apple's tooling for structured output via Swift types, enabling reliable, typed AI subsystems. The piece calls for a shift back to local-first software, using cloud models only when genuinely necessary.
Key facts
- Developers often add AI features via API calls to OpenAI or Anthropic.
- Cloud-dependent apps break when servers crash or credit cards expire.
- Local devices have powerful Neural Engines that are mostly idle.
- Streaming user content to third-party AI providers raises data retention, consent, and privacy issues.
- The Brutalist Report iOS app generates summaries on-device using Apple's local model APIs.
- Apple provides tools for structured AI output via Swift structs with typed fields.
- Local models are excellent for summarization, classification, extraction, rewriting, and normalization.
- The author urges using cloud models only when necessary and treating AI as a trustworthy subsystem.
Entities
Institutions
- OpenAI
- Anthropic
- Apple
- Brutalist Report
Locations
- Virginia