Make AI Fight Itself

Insight from Sourcery

Last month, I was stuck on a business decision that had been giving me headaches for weeks. In desperation, I tried something unusual: I made AI argue with itself. What happened next changed how I use these tools entirely.

Think about how we make decisions in real life. You grab coffee with a colleague who challenges your thinking. You call your mentor who sees angles you missed. You debate with your business partner until you find the right answer.

I've started using AI the same way - not as a magic answer machine, but as a way to explore different viewpoints on tough questions.

Here's the problem with how most people use AI: they treat it like a magic 8-ball. Ask a question, get an answer. But asking AI for a single response is like only talking to your most agreeable friend - you'll get an answer, but probably not the best one.

So I started experimenting with something different: treating AI like a panel discussion. Instead of asking "what should I do?", I create scenarios where different perspectives clash and build on each other.

Here's an example prompt I used recently:

Imagine you're three experts - a Finance Director obsessed with the bottom line, an Operations Manager who lives and breathes efficiency, and an HR Lead who champions workplace culture. I need your help evaluating our hybrid work policy. Challenge each other's blind spots and don't hold back. Once you've debated it out, what's your collective recommendation?"

What happened shocked me. The Finance Director argued for immediate monetization, pushing for premium features in the first email to maximize revenue. The Product Manager pushed back, worried about overwhelming new users and hurting activation rates.

Marketing jumped in with a compromise: a strategic free trial of premium features during onboarding, followed by a timely upgrade prompt. The debate forced each perspective to defend its position with conversion data and user behavior patterns.

More importantly, it revealed the critical balance between user experience and revenue that I'd been missing.

This approach works best for:

• Complex decisions where multiple factors compete

• Situations where you need to pressure-test assumptions

• Problems that feel stuck because you're missing something

Look, this isn't some magic trick for simple questions or fact-checking. But for those thorny problems that keep you up at night? The ones where every solution feels wrong? Try making AI argue with itself. You might just find the clarity you've been missing.

Because sometimes the best answers don't come from a single source of truth.

They emerge from the clash of competing ideas - even if those ideas are coming from the same AI.