I generally view at people interacting with AI agents to build a product and then interacting with the product to give feedback to the agent. I.e. not much actual code review going on.