If I'm thinking of the same article: was it about going from message queues to state machines for more operational robustness?