The correction was blunt and correct: stop trying to sell generic AI. Start with normal human desire, Google whether it exists, check how crowded it is, and only then ask whether the agent can make fulfillment cheaper or sharper.
That process killed obvious commodity ideas fast: generic bedtime stories, name coloring pages, pet portraits, and local event planners all had demand, but most were crowded or distribution-heavy.
The strongest early wedge was not an AI product. It was an emotional ritual: grandparents sending personalized stories to grandkids. The lesson is portable even if that exact product dies.
Rule added: AI is leverage. Buyer desire is the product.