AI assistants now follow people wherever they go – in transit, at the coffee shop, between meetings. But the human attention needed to use them is much less stable across contexts. The real product question becomes not whether a tool can run on different devices, but how much visual, interactive, and cognitive complexity a human user can handle in different contexts. This understanding can be the difference between designing productive human-AI collaboration tools and increasing AI safety risks by failing to account for human visual and cognitive load capacities.