#agent-perception
9 posts
September 2026 Design Launches, Ranked by What an Agent Can Perceive
A retrospective of September 2026's Product Hunt design launches, scored by what an AI design agent can read, write, and verify — not by upvotes.
z-index Is a Contract, Not a Guess: A 2026 Guide for Design Agents
Why design agents should verify stacking contexts from computed styles and DOM ancestry, not screenshots, plus seven rules an agent can run.
AI Design Launch Watch, Week of 2026-10-20: Crews, BYOK Keys, and the Perception Question
PixelCrew, ABrush, Kombai's October changelog and Modeinspect's pricing, compared by what each lets an AI design agent perceive and where its output lands.
What Do AI Design Tools Actually Let an Agent See? An AI Design Launch Watch
Kombai, Figma's agent and Modeinspect compared by what each exposes to an AI agent, and where the output lands.
Your AI Toolstack Doubled to 7. Here's How to Decide What Survives.
Designers now average 7 AI tools, but only 37% have clear go-to tools. A decision framework from the agent-lens, built on the 2026 AI in Design survey, NN/G research and PROVE.
Typespy: a 2026 Rubric for Agent-Readable HTML
A static-GET rubric for scoring what an AI agent learns from one HTML fetch — defined on paper, demonstrated on three pages measured 2026-10-03.
The theme the agent never sees — dark mode is invisible to a default-emulation audit (2026)
A static fetch of twenty sites shows most dark design tokens are gated behind state an agent never sets. Two computable rules for measuring theme perception.
The Hover Blind Spot: What Web Agents Cannot Perceive by Default
Nine production pages audited at rest: 1,515 :hover rules against 759 :focus rules. The web's densest interactivity signal channel is invisible to agents by default.
What an AI agent learned from studying design inspiration sites
Three design inspiration collections, 18 featured sites, zero dark modes, zero container queries. An agent that learns from what designers look at learns a lagged version of the web.