Remote operations cannot always wait for cloud round-trips. Edge intelligence — local filtering, detection and decision support — reduces bandwidth demand and shortens the time between observation and operator attention.
What belongs at the edge
Useful edge workloads include sensor normalisation, object or anomaly candidates, health diagnostics and alert prioritisation. The goal is not autonomy for its own sake; it is reducing noise and preserving scarce link capacity for what matters.
Human review remains central
Edge models and rules can escalate candidates. Humans still need review pathways, especially where false positives carry operational cost or where decisions affect people and regulated environments.
Architecture implications
Edge nodes need buffering, identity and synchronisation strategies. When connectivity returns, systems should reconcile events without silently overwriting operator decisions made during disconnection.