535 humans and capable AI agents still left gains on the table

A study of 535 participants with real AI model capability scores found that giving humans access to capable AI did not, on its own, produce better outcomes. The assumption baked into most product design, that capability plus availability equals use, did not hold. Humans left the gains unclaimed. The finding matters for anyone designing the moment when a human passes work to an agent. Exposure to a capable agent is not the same as a human knowing when to pass work over, how much to hand over, or how to judge what came back. The interaction design between the human and the agent determines whether the capability gets used at all. For UX leadership, this study reframes the core design problem. The question is not whether the agent is capable enough. The question is whether the human has enough in view to decide to use it, shape what it does, and trust what it returns. Capability without a visible process, a legible result, and a clear moment of judgment stays unclaimed.