My new apps seem to be getting in to some kind of review / validation loop by an infinite eternal battle with the ‘code reviewer’. See attached. And agent seems to be adding very unnecessary functionality to a seemingly simple app. Goes on forever unless I stop it.
I haven’t encountered this in Agent 4 yet but in Agent 3, the ‘architect’ did at times go too far with recommendations. I think you have the be mediator or set the guardrails. These discussions usually ended with me asking for just the critical recommendations and outright rejecting the unnecessary ones (using that language to get the point across). The agents have a very large context windows so if you keep the chat going (instead of starting new ones too frequently), the agent should learn your preference for simple solutions. Many times I use that exact language in my prompts.