Taste Labs mined 2M sites: AI didn't start homogenization, it accelerated it

AI Engineer · youtube · 2026-09-11

Thais Castello Branco, founder of Taste Labs (out of stealth weeks before this talk), shares an analysis of 2M+ websites over a decade: the internet was homogenizing before AI — palettes and layouts converging, trends spreading faster. AI didn't start the collapse; it accelerated it and made it context-blind, so a pet shop and a finance firm end up with the same page.

She names three signatures of slop: repetition, lack of fit, low intent — and prefers "judgment" over "taste," rejecting the fix being everyone acquiring taste, since designers spend careers on it via exposure, pattern recognition and restraint. To make the fuzzy measurable: contrast, alignment and palette scoring become near-deterministic with tight context; aesthetics leans on data. Her team mined site features and trained "probes" — small classifiers each detecting one characteristic — whose combined frequency predicts slop, beating the ask-a-model-to-judge approach. On solutions, she argues inference time matters as much as the model, since intent and context are exchanged there: one product deliberately goes out of distribution (breaking chosen rules while respecting category expectations, rather than raising temperature); another turns a brand into structured components an agent can follow and be graded against.

Original post →

More from Apps

Apps channel →