Researcher proposes method to give AI agents subjective judgment—enabling them to make design choices and trade-offs without explicit instructions

Emil Kowal published a conceptual framework for embedding 'taste' (subjective aesthetic and practical judgment) into autonomous AI agents, addressing the gap between what agents can execute and what they should choose to do when multiple valid solutions exist.

Rather than hard-coding every decision rule, the approach lets agents develop preference patterns—similar to how a human designer learns to pick fonts, layouts, or trade-offs between speed and beauty—allowing them to operate with implicit values instead of explicit instructions for every choice.

This matters to product teams and designers: instead of writing detailed specifications for every UI decision or design detail, you could give an AI agent a taste profile ('clean and minimalist' or 'high-contrast and accessible') and let it autonomously generate interfaces that match your brand without constant human correction.

The work is currently conceptual and available as a blog post; no released tool or product exists yet, though the idea directly addresses why current AI agents often produce sterile, over-engineered outputs rather than naturally cohesive designs.