The algorithmic era was constructive by nature: you began with basic design elements and complexity was built up through iteration and combination. The designer’s skill was compositional: knowing how simple operations compound into rich systems.

Generative AI flips the ontology of the starting point. The model doesn’t give you primitives; it gives you output that arrives already saturated, dense, over-complete, drawn from a latent space that contains everything at once. So the designer turns from construction to negation.

Prompting itself is largely subtractive — you’re not building an image, you’re carving away the enormous default space of what the model would produce. It’s the old distinction between modeling in clay and carving in marble, except now everyone works in marble whether they want to or not.