Kill Your Darlings: Why I Built 4 AI Agents to Attack My Own Ideas

Kim Wong

— 1 min read

I built 4 AI agents whose only job is to destroy my ideas.

Not refine them. Not validate them. Destroy them.

Here's why.

When I have a new business idea, my instinct — like everyone's — is to look for reasons it will work. Confirmation bias kicks in before I even finish the sentence.

So I flipped the process. I built 4 adversarial agents, each attacking from a different angle:

  • The Sharpener — forces the idea down to ONE use case. No "platform for everything" allowed.
  • The Reality Checker — can we actually build it? Is it safe? Will anyone come back on Day 7?
  • The Destroyer — tries to kill it outright. Finds the existing competitors you missed. Finds the legal landmine you ignored.
  • Market & Money — how do you reach users with zero budget and no network? If the answer is "go viral," the idea is dead.

They run simultaneously. They don't coordinate. They don't care about each other's feelings.

Then I synthesize their attacks with my own judgment. Not blindly — agents drift, overstate threats, and sometimes hallucinate competitors that don't exist. So there's a 5th agent (a fact-checker) that audits the claims.

The results so far: 7 ideas in. 3 killed in Round 1.

That's not failure. That's the system working. Each kill saved me months of building something nobody needed.

Here's the thing — this approach isn't new. It's ancient:

  • The Catholic Church appointed a Devil's Advocate (literally) to argue against sainthood candidates. For centuries.
  • The military uses Red Teams — dedicated units whose job is to defeat their own side's strategy before the enemy does.
  • Gary Klein's Pre-mortem: "Imagine the project has failed. Now tell me why." Used at every serious org from Amazon to the CIA.
  • Darwin kept a "golden rule": every time he found evidence against his theory, he wrote it down immediately — because he knew his brain would try to forget it.

The principle is the same across every era: the best way to strengthen an idea is to try to kill it first.

AI just lets you do it in 10 minutes instead of 10 meetings.

If your idea survives 4 adversaries arguing in parallel with real market data — it might actually be worth building.

If it doesn't? You just saved yourself the most expensive lesson in entrepreneurship: building something nobody wants.

Read more