Most people use AI to confirm what they already think. That’s backwards.
When I’m reviewing code, drafting a proposal, or planning a project approach, I ask Claude to argue against my solution. Poke holes. Find the edge cases I missed.
It’s not about being negative. It’s about catching problems before they become expensive.
I take it a step further: I pit different LLMs against each other.
Get one opinion from Claude. Get another from ChatGPT. Run it by Gemini. When they agree, you’re probably on solid ground. When they disagree, that’s where the interesting questions live. That friction is valuable - it highlights assumptions you didn’t know you were making.
No single AI is going to be right every time. But a panel of devil’s advocates? That’s getting closer to something useful.
Try it next time you’re confident about something. You might be right. But you’ll know why you’re right.