Agents with Taste
The author explains how to turn design decisions—like initial scale, easing, and typography rules—into explicit “skill files” that coding agents can follow. By articulating why certain animations feel natural, you guide AI agents to produce consistent, high-quality results. He demonstrates using Anthropic’s skill-creator and Claude Code, and shares a public skill package.
- Coding agents produce mediocre visual/animation work because they lack a trained eye, but this can be fixed by encoding design taste into explicit "skill files" with concrete rules instead of vague preferences
- Concrete example: starting animations at scale(0.85) instead of scale(0) creates natural-feeling growth versus jarring pop-in—showing how gut-feel taste can become codified logic (same applies to easing curves via decision trees)
- Using Anthropic's skill-creator, Claude Code can review existing animations against these rules and output specific violations plus before-and-after code fixes
- All these rules are packaged into an installable skill via `npx skills add emilkowalski/skill`
Coding agents are good at structure and logic but tend to miss what makes visual work feel right. Emil Kowalski's fix is to write down his own judgment as skill files, which are rule documents an agent can follow. His example: an animation that scales from `scale(0)` looks wrong because the element seems to come out of nowhere. Starting from a higher value, something like a balloon that never fully deflates, reads as more natural and gentle. The point is that these "taste" calls usually have a reason behind them, and once you can state the reason, you can hand it to an agent as a rule instead of hoping it guesses correctly.
He applies the same approach across the animation pipeline. One skill file holds practical tips like the scale trick. Another has a strict easing flowchart so the agent picks an easing from his stated philosophy rather than inventing one. Duration gets its own rules, and he has similar files for layout, icons, color, and typography. His advice for building one is to ask why you made a given decision, write the reasoning down clearly, set the rules, and be strict about them. He uses Anthropic's skill-creator skill to help package this. In his demo, he asks Claude Code to improve a dialog animation under his skill, and the agent returns a list of issues tied to his rules plus a before-and-after table of changes.
The concrete payoff is a sharper, more specific workflow: your judgment, written as rules, applied across every component an agent touches. He has bundled his own blog posts into a design engineering skill, installable with `npx skills add emilkowalski/skill`, covering animation, component design, and principles from his open source projects such as Sonner. The article is also a pitch for his animation course, which had two days left to enroll when the page was captured.