Does perceived control matter more than actual control for AI trust?

Gunasekaran argues that users of agentic AI systems do not need constant, hands-on control to feel safe — they need the sense that they could intervene if something went wrong. She writes: "Users do not require constant control to feel safe. They require perceived control. The difference between these is subtle but profound." A system that permits intervention, even when the user rarely exercises it, "feels trustworthy," while one the user "cannot interrupt, override, or question quickly becomes threatening, regardless of how accurate or efficient it may be."

Her reasoning is psychological rather than technical: "humans measure safety by their ability to intervene when something feels wrong." The design implication she draws is that UX for agentic AI must "design the experience of control as carefully as we design an AI's autonomy" — control becomes a felt, perceptible property of the interface, not merely a backend permission or kill switch that exists but stays invisible to the user. This reframes human control as a UX design object in its own right, distinct from how much control a user actually ends up exercising.

This sharpens the control dimension that recurs across adjacent frameworks. The What UX principles do workplace users want in AI agents? study finds users weight human control heavily but does not explain why a rarely-used override still matters; Gunasekaran's psychological framing supplies that mechanism — it is the felt availability of intervention, not its frequency, that produces trust. It also complicates Can automation and human control increase together?: where Shneiderman treats control as a structural axis to be maximized alongside automation, Gunasekaran treats control as a perception to be designed, which could in principle be engineered even where the actual intervention capability behind it is limited.

The essay does not test this claim empirically — it offers no study, user research, or data on what specifically makes control "feel" available, nor how designers might measure perceived versus actual control, or where the two diverge in ways that mislead users (a sense of control without the real capacity to intervene could be comforting but false). The piece is a practitioner argument, not a validated finding; it points to a hypothesis designers might test rather than a settled design principle.