how i work

Most of it comes down to one habit: finding out what a screen is obeying before I redesign it. The rest is how I use AI without skipping the thinking, how I work with the people around me, and how I keep accessibility in from the start rather than bolting it on at the end.

How I make product decisions

The same thing was true in all three case studies on this site. The interface was doing exactly what the system underneath told it to, and the person using it was the one paying for that. So before I redesign a screen, I find out what the screen is obeying.

Read the system first

01

When a flow is confusing, the cause is usually upstream: an order that mirrors how data is stored, or a step that exists to serve an internal process. I find out what a screen is obeying before I change it.

Get evidence, even without analytics

02

Not every product is instrumented, and support tickets alone are not proof. Where there is no data I triangulate support contacts, surveys and interviews. Where testing is possible, I run it before proposing a direction.

Name what it costs

03

Every decision costs something, and I say what. Backend work instead of a cheap interface patch, a step that lands in a system I did not design, complexity hidden from the user rather than removed. A trade-off that is not stated has not been made.

Read the system first

When a flow is confusing, the cause is usually upstream: an order that mirrors how data is stored, or a step that exists to serve an internal process. I find out what a screen is obeying before I change it.

01

02

Get evidence, even without analytics

Not every product is instrumented, and support tickets alone are not proof. Where there is no data I triangulate support contacts, surveys and interviews. Where testing is possible, I run it before proposing a direction.

03

Name what it costs

Every decision costs something, and I say what. Backend work instead of a cheap interface patch, a step that lands in a system I did not design, complexity hidden from the user rather than removed. A trade-off that is not stated has not been made.

Collaboration

Design decisions rarely land the first time I propose them. I push back with reasoning I can back up, and I stay involved past handoff instead of treating a build as someone else's problem.

Icon

UX Office Hours

A recurring design critique I started and run at Ruby. Open across the team, with POs and developers regularly in the room, and its own channel for planning sessions and interim testing between them.

Icon

Feedback rounds

Individual pieces of work also go through their own rounds with POs and developers, so decisions get tested before they ship, not after.

Icon

QA and tracking

My involvement doesn't stop at handoff. I stay close to QA on what I design, and I'm involved in how it gets tracked once it's live.

AI workflow

AI speeds up every stage of this process, from desk research to documentation. But every output goes through me first. I challenge one model's answer with the other, ChatGPT against Claude, and keep going until the reasoning survives both. Nothing moves forward until I've reviewed and validated it, so the thinking stays mine even when the drafting is faster.

01

Desk research and benchmarking

Exploring patterns, existing solutions and questions worth investigating before I commit to a direction.

ToolsChatGPT
02

Research synthesis

Structuring large amounts of information and pulling out themes. I always validate this against the original research. An AI summary is not evidence on its own.

ToolsChatGPT
03

Ideation

Challenging my own assumptions and forcing a look at flows I wouldn't have generated alone.

ToolsGoogle StitchClaude Design
04

Prototyping

AI-assisted development to build something people can actually use, not just look at.

ToolsClaudeLovable
05

Documentation and knowledge sharing

Turning research, decisions and product logic into something the wider team can pick up and run with. I connect Claude to Notion through MCP, so documentation lands where the team already works.

ToolsClaudeNotion AI

Accessibility compliance

Accessibility isn't something I bolt on before launch. The goal is a product that actually works for people using assistive technology, not just one that passes a checklist.

project image

A loop, not a launch checklist

I test against WCAG criteria and map every issue by severity, fix in priority order starting with what blocks people the most and keep reviewing after the product ships.

01

Audit

02

Fix by severity

03

Ongoing review

project image

A loop, not a launch checklist

I test against WCAG criteria and map every issue by severity, fix in priority order starting with what blocks people the most and keep reviewing after the product ships.

01

Audit

02

Fix by severity

03

Ongoing review