I've adapted my Special Theory of Autonomous Agent Reliability (STAAR) to predict the reliability of LinkedIn posts about "AI". In STAAR, the reliability of a single step - a single model interaction - is given as: R = 1 - (1 - C)(1 - P) Where C is the probability the model's output is correct, and P is the probability that any errors are caught before they propagate and compound. So the reliability of N steps in a workflow is Rᴺ.