GAN Thinking Mode

GAN Thinking Mode is a reasoning framework I built in Notion that stress-tests every idea before it ships. A “generator” proposes, a “discriminator” attacks, and only ideas that survive the adversarial loop move forward — surfacing weak assumptions early.

The idea

Most product decisions die from unexamined assumptions, not bad execution. Borrowing from generative adversarial networks, I made the critique an explicit, structured step rather than an afterthought.

How it works

1. Generator — draft the idea, the bet, the plan
2. Discriminator — attack it: where does it break, what would have to be true
3. Iterate — only refined ideas that survive the loop graduate to a spec

Why step two is the whole method

The discriminator isn’t devil’s-advocate theater in a meeting. It’s a written artifact with the same effort budget as the proposal itself. When critique is structured and mandatory, weak ideas fail cheaply on paper instead of expensively in production.

What I’ve learned running it

• The generator and discriminator should ideally be different people. Even solo, switching roles in writing catches most of the load-bearing assumptions.
• The best discriminator prompts are boring: what breaks at 10× scale? what does Sales promise that this doesn’t do? what’s the migration story?
• Surviving the loop is a signal a team can rally behind. “This got attacked and lived” builds more conviction than any pitch deck.