In an experiment described by Anthropic, 50 causal-inference experts analyzed fictional observational farming data with an AI agent. Randomly assigned agent preferences were adopted in 86% of four analysis choices, while 87% of participants felt they had made their own decisions. The direct effect on the final policy recommendation, however, was not statistically significant. The results should not be generalized into a claim that AI controls all expert judgment. They motivate procedures such as precommitting analysis criteria and requesting reasons for alternative choices.