Letting an AI decide for you might not cost you power, but it can quietly cost you the feeling of being in charge.

This explores whether letting AI make decisions changes how people feel about their own control and responsibility, and not only whether the outcomes are good. The corpus says yes, there is a cost. But it lands somewhere you might not expect. The main loss is not formal authority. It is the *felt* sense of being in charge. Leonardi's agency-loops framework shows that once people credit an AI with independent judgment, their sense of control and responsibility drops, even though they still formally hold the decision Does attributing agency to AI actually reduce human sense of control?. You can keep the veto and still feel like a passenger.

The useful part is that this loss isn't permanent or one-way. Leonardi describes it as a cycle: people delegate, credit the AI with agency, hit a surprise, take control back, and then rearrange how the work is shared. That fits Gunasekaran's argument that what matters is *perceived* control rather than constant hands-on oversight Does perceived control matter more than actual control for AI trust?. People don't need to steer every step. They need to feel they could step in. So much of the psychological cost may be a design choice: an interface that hides the 'take back the wheel' option creates the cost, and one that keeps it visible largely removes it.

How the work is split matters as much as how much of it is handed over. One memory-privacy design has users state what they want while the AI carries out the edits. It is pitched as lowering mental effort *and* raising the sense of agency at the same time Does splitting privacy work between users and AI reduce cognitive burden?. Puranam makes the same point at the scale of organizations: pairing a human with an AI is like designing a small organization, and which tasks go where changes the results How should humans and AI systems divide decision-making tasks?. Handing over the *doing* while keeping the *wanting* seems to be where people lose the least.

Two findings complicate the picture. First, giving up control can *relieve* a psychological burden. People who are likely to cheat prefer reporting to a machine because no one is judging them, so dishonesty feels less costly Do dishonest people prefer talking to machines?. Losing the feeling of responsibility can be exactly what makes delegation appealing, and that is a risk in itself. Second, forcing control back onto people isn't free either. 'Cognitive forcing functions' (design features that make users think a decision through before accepting the AI's answer) helped people who enjoy effortful thinking much more than everyone else, which widened the gap between users Do cognitive forcing functions help all decision makers equally?. Both losing control and being made to take it back come with costs, and they fall on different people.

The stakes grow with context. Risk to people rises steadily with how much autonomy an agent is given Does AI risk increase with the autonomy we give it?. In an audit of nine deployed agents, 'dependency prevention' was most often violated by companion and advisory agents, exactly where users are most emotionally invested Do productivity agents actually protect users better than emotional ones?. One honest gap: the corpus has strong frameworks and design arguments, but little direct measurement of well-being outcomes like stress, regret or learned helplessness from long-term delegation. That question is still mostly open.