Kinda an obvious take but as i keep heavily using Deep Research + Agent + o3 + o3 pro There's a very clear line between the things you are OK if an agent goes off to do (out of flow state) vs things you want a model to get back to you on within 10 seconds (stay in flow). I'd imagine there's some user steering that can happen here and also product experiences to build around this pretty hard line of in-flow AI vs. out-of-flow AI. Currently o3 is for in-flow, o3-pro pseudo flow, DR daily flow (so, out of flow mostly), Agent completely out of flow.