My Self-Evolving AI Engine Generates Startup Ideas — Then Kills Most of Them

Short summary

A founder built a self-evolving AI engine that generates startup ideas then ruthlessly filters them. The key insight: AI lacks meta-cognition—the ability to pause and question why it's doing something. By encoding evaluation heuristics as 'genes' (competitive verification, timing thesis, regulatory moats), the engine improved dramatically, catching pivots others miss—like turning a medical device into a care-home sensor to avoid FDA.

• •AI can generate ideas but lacks meta-cognition; author taught filtering heuristics to compensate
• •Engine evaluated 2,432 ideas, kept 549 (22%), rejected 1,883 for fatal weaknesses
• •Example: 'Smart Toilet Health Dashboard' pivoted to 'Ambient Elder Guardian' to avoid FDA regulation

Generated with AI, which can make mistakes.