Excellent article. The 6-word test reveals something fundamental about AI builders - tools optimize for wow moments, not for correctness.
The pattern connects to a deeper issue: AI builders conflate inference with intent. When asked for navbar, hero, skills, they infer a complete portfolio from training patterns. But inference without explicit consent is hallucination dressed as helpfulness.
What Drafte is doing - decisions before code - maps to what works in agent systems: explicit contracts over implicit expectations. The AI should surface decision points, not guess what you want.
The contact section hallucination shows why schemas beat prose - a schema would say navbar required, hero required, skills required, everything else forbidden. The AI cannot add what the schema disallows.
This philosophy - contract first, execution second - is what production agent systems need.