Local-first AI as a Product Constraint
A sample research note on how on-device models change privacy, latency, resilience and interface decisions.
The constraint changes the product
Local-first AI is more than a deployment choice. When inference happens on the device, privacy, availability and latency become properties of the product architecture rather than promises added around a remote service.
Capability is not uniform
Device models introduce a different kind of progressive enhancement. The interface needs to remain useful when a model is unavailable, warming up, or less capable than a hosted alternative. Designing that fallback deliberately is part of the product work.