Benji TaylorThe gap between seeing and describing.
Summary
Visual AI tools let you point straight at an element but lock you into one editor, while flexible tools like Claude Code make you describe everything in words. Benji's point is that describing loses precision exactly where you need it. The button hover feels sluggish, but which part, the delay or the duration or the easing? Screenshots only help a little, because the agent still has to guess which part you mean. The gap hurts most on animation, where the state you want to talk about lasts a fraction of a second. His answer is a localhost overlay that captures selectors, positions and context alongside the note, then emits markdown any agent can act on.
View full writingGoes well with
emil-design-eng
The design-engineering house style behind Sonner and Vaul.
Emil's main skill. Mostly animation, with the UI-polish and component judgement that surrounds it: what to build, how far to push it, and when to stop.
npx skills add https://github.com/emilkowalski/skills
review-animations
Reviews your motion strictly, against Emil's rules.
A production-critical pass over animation quality, timing and consistency in a diff. Pairs with improve-animations, which does the same for a whole codebase.
npx skills add https://github.com/emilkowalski/skills --skill review-animations
improve-animations
Audits every animation in a codebase and hands back prioritised plans.
Read-only on source. Surveys the motion you already ship and produces self-contained implementation plans any agent can execute, ordered by what actually matters.
npx skills add https://github.com/emilkowalski/skills --skill improve-animations
Emil Kowalski