AI workflow

How I design with AI

I use AI to get to testable ideas earlier. It helps me research, synthesise, prototype and document. The problem, the decisions, accessibility and quality stay with me.

2026

What changed in 2026

In 2026, the agency I work with moved its design workflow from Figma to Claude Code. After years designing in Figma, I now also design directly in code. On Apta, that meant meeting the client's brief of a live beta in 2 weeks. On Colleago, I work in the codebase alongside two other designers and the engineering team.

2 weeksApta, from kickoff to a live beta
~10 daysfrom concept to testable build, compared with roughly 12 weeks on a similar earlier design-and-handoff cycle
The workflow

Where AI helps, and where I decide

Discover

AI helps withDrafting market scans, competitor reviews and early personas.

I decideWho we're designing for and what's true. On Apta, I checked every AI-drafted claim against product sites and app store listings, and tagged anything unconfirmed as an assumption.

Align

AI helps withTurning workshop boards into a written summary and first-draft PRD.

I decideThe problem statement, the scope and who owns each action. The workshop itself is always human.

Prototype

AI helps withBuilding screens in code from real design system components, guided by a CLAUDE.md file of product rules: who it's for, what it must never say, design tokens and voice.

I decideThe flows, interaction quality, accessibility and craft. Every review is a working build, not a mock-up.

Test

AI helps withDrafting test scripts, and turning session recordings into a first draft of findings.

I decideWhat the findings mean and what we change. On Apta, the team shipped fixes for all 5 usability issues within a week of testing.

Document

AI helps withFirst drafts of PRDs, specs and decision logs.

I decideWhat's accurate, what's decided and who's accountable.

CLAUDE.md product context file for Apta
The product rules Claude Code reads before building any Apta screen: who it's for, what it must never say, tokens and voice.
Working in a real codebase

Designing in code, safely

Designing in a real codebase means working the way engineers do. Early on at Colleago, design changes went through CI straight to main without a review, and a branch I'd cut from an older version overwrote another designer's component update. We only noticed once it was live. Since then, I branch from the latest main, and every change goes through a pull request reviewed by a designer and a developer. It's slower by a few minutes and safer by a lot.

Human judgement

Where I stay accountable

The problem

AI can suggest answers. Framing the right question stays with me.

The decisions

Every trade-off has a person behind it, and on my projects that's me and the team.

Accessibility

I check text size, tap targets, contrast and small-screen layouts myself. On Apta: 44px tap targets, 16px minimum text, and every screen checked at 360px wide.

Quality

Nothing reaches users until I've reviewed it for product intent, accessibility and quality.

Toolkit

Tools I use