Great finding! What is your approach to tackle the indeterministic outcomes of AI agents? On a side I am playing around with multiple agents verifying each other’s output.