366-person study finds AI lifts confidence faster than skill

A controlled study of 366 participants found that working with AI raises confidence more than it raises performance. Humans who used AI help felt more certain of their answers, but their accuracy did not rise to match that certainty. The gap between how confident a person feels and how well they perform grew wider with AI assistance. This matters for anyone designing a product where a human reviews, approves, or redirects agent work. A person who feels certain is less likely to question what the agent produced. They skim instead of read. They approve instead of inspect. The interface does nothing wrong, but the human's sense of mastery has outrun their grasp of the work. An agent that produces polished output gives the human a signal that everything is fine, even when it is not. Designers who want humans to stay engaged with agent output need to think about how the interface surfaces uncertainty, invites skepticism, and keeps the human responsible for the judgment, not the approval alone.