The escalation path — not the opposite of the edge. When a task exceeds the device, the local model, or an app's privacy boundary, the discipline is knowing when reaching past it is justified, how it is constrained, and what returns to the record.
An app can run a capable model on device, and most of the time it should. Cloud Intelligence is the discipline of the exception: the moment a task exceeds the device, the local model, or the privacy boundary a specific app is willing to hold. Apple's own framing makes the boundary explicit — the same Swift API can resolve to an on-device model, to Apple's models on Private Cloud Compute, or to a server provider.
The goal is not to send everything outward. It is to know, for a given task, when reaching past the device is justified, how the escalation is constrained, and what returns to the system of record. Escalation is a decision with a cost and a boundary — and therefore a governance decision.
On device by default. The cloud as the exception you can justify, constrain, and audit.
Edge Intelligence asks where intelligence should run. Cloud Intelligence asks when that answer changes — and Governed Intelligence decides who sets the criteria, what is constrained, and what is logged when it does.
The essay, the build, and the field notes arrive in dedicated sessions. This page states what is coming rather than pretending it is already here.
On device by default; the cloud as the justified exception. The boundary is the work.
Cloud is not the opposite of the edge. It is where a task goes when it exceeds the device, the local model, or an app's privacy boundary.
Edge Intelligence →Escalation is a governance decision: who defines the criteria, what is constrained, and what returns to the audit log when intelligence reaches outward.
Governed Intelligence →Private Cloud Compute and server models extend the same substrate the lab runs locally. The boundary case of the workbench.
Silicon →