Ambarella’s Pietro Antonio Cicalese examines how natural-language prompts and multi-modal reasoning on power-efficient SoCs can enable contextual, site-specific operational intelligence even on PoE-powered edge cameras, addressing a “personalization gap” in physical security, where hard-coded analytics scale poorly and context has historically been configured manually per camera and per site through complex VMS interfaces and task-specific analytics modules. In its 2026 State of Cybersecurity report, Ivanti reported that 87 percent of security professionals are now prioritizing agentic AI adoption. At ISC West, conversation shifted from awareness to deployment mechanics and how systems should adapt behavior to site-specific operational requirements with minimal manual configuration; several vendors demonstrated natural-language video search, custom alert creation, and on-device AI running entirely at the edge. ONVIF reported that conversations were no longer on if AI belongs in physical security, but on what standards conformance and interoperability need to be in place and the practicalities of deploying AI-enabled and cloud-connected systems across multi-manufacturer environments. Early AI-enabled video surveillance used fixed-function analytics such as motion detection, line crossing, and basic object classification, which were effective but strictly limited, since adapting a narrowly trained model requires another project, integration cycle, and maintenance burden. An “app store” model also failed because physical security deployments are selected at installation and rarely revisited, leaving no economic incentive to maintain, curate, and update a catalog across a fragmented ecosystem of camera OEMs, VMS