Nick Perry shows how to make Claude debate itself before answering. Set up internal agents that hunt for flaws, question your framing, and push for next steps. They critique each other until a final agent decides, turning Claude into a pressure-tested advisor instead of a yes-man.
The best AI users don't ask it for answers anymore. They make it argue with itself first. And the reason why isn't shocking at all. See, instead of getting one answer from Claude, your idea goes to a panel inside Claude where there are multiple agents who argue about the best answer. One hunts for the fatal flaw, one asks if you're even solving the right problem, and one only cares what you should do next. Then they read each other's answers, tear apart the weak arguments, and then one makes the final call. So instead of Claude being a yes man and telling you your ideas are genius, it pressure tests them before you bet thousands of dollars in 3 weeks on the wrong thing. And that's why—
