@Jeremiah i recently saw your talk that you gave here you mentioned its not a great idea to have agents go into state machine like loops because of hallucinations. is this the case even with clearly defined parameters via function definitions and guard rails on 4o? ReAct is essentially a similar loop which is the current industry standard to let an agent be "autonomous". would like to know your opinion on this specially since it doesn't seem easy to get controlflow agents to behave like ReAct agents (which i think is not an incorrect choice)