Much of the current conversation around intelligent software treats generation as the hard part. That can be useful. It is also an incomplete picture of how real work actually improves.
Businesses and creators often already have years of customer language, unfinished ideas, correspondence, drafts, analytics, and lived knowledge sitting in disconnected systems.
Generation is cheap. Context is scarce.
When software begins from a blank prompt, it asks the person to recreate the world. That recreation is expensive—and unreliable.
Discovery software should start with what already exists.