AI design practice · Arity

AI design as a craft, not a shortcut.

I treat AI tooling like any other design surface: it deserves a system, a brand voice, and a real onboarding story. This page is a tour through what I've built across Arity — the pieces other teams plug into.

  • Days → hours Concepting compression Early-stage concepting is now a sprint, not a week.
  • Non-designers Upskilled into super users Marketing colleagues self-serving lightweight creative work.
  • 1 source Of brand voice truth Reused by every AI workflow that touches user-facing language.
01

Pattern library

An AI-friendly Figma pattern library

Most pattern libraries are organized for humans. I architected ours so AI tools could read it cleanly — consistent naming conventions, tokenized properties, rich annotations, predictable variant structures. The output quality from AI tooling jumped meaningfully once the library spoke a language the model could parse.

02

User guides

Cursor user guides for design and product teams

Practical, opinionated guides covering prompt patterns, when to reach for Cursor vs. when to stay in Figma, how to integrate AI output with our existing design system, and the failure modes to watch out for.

03

Skills · Tone & voice

Brand voice as shared infrastructure

Tone, voice, and brand-alignment skills (and Cursor rules) so AI-generated copy and design output consistently sound like the brand. One source of truth, called by every workflow that touches user-facing language.

04

Onboarding

Turning non-designers into super users

Cross-functional AI onboarding for marketing colleagues — moving them from "I tried this once" to capable self-serve users of the AI design stack. The goal isn't to replace design; it's to free design capacity for higher-order strategy.

05

Sprints · Rhythm

AI-assisted design sprints

A repeatable sprint format that compresses early-stage concepting from days to hours: a scoped problem, a structured set of prompts, a critique loop, and a clear handoff back into the normal design process.

Sample artifact

A snippet of the brand voice skill.

Real skills are longer and more specific — but the shape is the same. A skill is a small, opinionated contract between the team and the model.

brand-voice.skill.md md 
---
name: brand-voice
description: Author copy in the Arity Perceptive Advisor voice.
---

# Voice principles
- Lead with the user benefit, not the system feature.
- Surface confidence — never imply false certainty.
- Cite the data behind a claim where useful.
- Choose warmth over hype, evidence over jargon.

# Avoid
- Finger-wagging or moralizing about driver behavior.
- Buzzwords ("AI-powered", "revolutionary", "seamless").
- Passive constructions that hide who is doing what.

# Prefer
- Plain, specific language a friend would use.
- Sentences a tired user can parse on a small screen.
- Clear, unhedged "we" / "you" attribution.

# When generating UI strings
1. Start from the user's job, not the screen's job.
2. Surface confidence levels honestly ("usually", "based on...").
3. Offer one focused next step, not a wall of options.

# Examples
- ✗ "Your driving was suboptimal this week."
- ✓ "Two harder brakes on Tuesday — usually quieter than this."

Want the full case

There's a longer story behind this.

The Cursor AI case study walks through how the practice came together — what worked, what didn't, and what's next.

Read the case study