Agentic AI boosts physician trust 89% in clinical decisions

Medical AI is evolving from static reasoning to agentic AI—a paradigm where systems autonomously invoke external tools and render intermediate reasoning steps transparent to users. A new study published on arXiv (accepted at AMIA 2026) explored whether this transparency actually builds physician trust. Three physicians evaluated 315 multimodal clinical cases, comparing agentic AI against non-agentic baselines. The study measured both process-oriented cognitive trust (how much physicians believed the AI's reasoning) and outcome-oriented behavioral reliance (whether they followed the AI's recommendations).

Results showed a dramatic preference: physicians exhibited significantly higher cognitive and behavioral trust for the agentic model (P < 0.001). In treatment planning tasks, physicians chose the agentic AI's reasoning in 89.57% of cases. Importantly, process-oriented trust was strongly associated with outcome-oriented reliance, suggesting that showing work builds confidence. Yet the study also found measurable over-reliance on incorrect agentic outputs, indicating that decision-logic transparency alone is insufficient. The authors emphasize the continued need for rigorous clinician oversight, highlighting a key limitation of even the most transparent AI systems.