It’s like herding cats, if you’re doing it right. There should be enough adversarial tension in your process that the majority stays on track, but the parts where you are managing and have strong observability are the points where AI needs supervision to not go wildly off the rails, and where you can cull or redirect pathological sessions.
We're seeing this everywhere now, people generating thousands of lines of documentation or PR feedback in seconds, then dumping the actual labor of verification onto their teammates.