Anthropic: We conducted an experiment on analysts and found that 86% of the time they tend to their AI agent's answer if they haven't expressed an a priori judgment, and if they have, then 60% of the time they change their point of view if the agent disagrees. Problem Statement: Questions with two "logical" options. In this case, the agents did not actually choose, but justified a random choice. On the one hand, there are questions about design definition: more like UX for coordination than for decision making. On the other hand, we are facing the same crisis of critical thinking, when artificial intelligence is the driver for the simplification (and flattening) of the decision-making model. In fact, we are rapidly moving to a state where human behavior can be predicted by the responses of its model. The same goes for organizational decisions. If you know your competitor has AI on Claude, then go ask Claude what he would advise in this or that situation. And you get the same "86%" probability that a competitor will do exactly that.