#machine-learning
4 posts
What Resolution Does an AI Design Agent Actually See?
A walkthrough of how Claude, GPT-5.6, and Gemini resize and tile screenshots — and how to compute the smallest UI element a design agent can actually perceive.
When to Fine-Tune a Design Agent — A Routing Rule for Hosted, Tunable, and Open-Weight Models
A decision framework for choosing between hosted prompting, managed fine-tuning, and open-weight deployment when teaching an AI agent to work inside a specific design system.
Taste Is a Training Signal, Not a Vibe
How Midjourney, FLUX, and Ideogram turned human preference into design-agent feedback loops — and why Design Arena's $7.9M raise proves taste is infrastructure.
The Medium Grid — Mapping 15 AI Design Models by What Agents Can Perceive and Verify
An agent can only learn from output it can parse, and only critique what it can perceive. Fifteen AI design models mapped on two axes — perception modality and output medium — with a routing rule for picking the right model per design task.