Process / Design with AI

Design with AI

I use AI to make design faster to validate — not faster to guess. The goal stays the same: the right product, tested with real people. The loops just get much shorter.

01Frame

A sharp problem, a hypothesis and the riskiest assumption to test.

02Prototype

A working, coded prototype with real data — in hours, not weeks.

03Validate

Real users and stakeholders on the real thing, the same day.

04Decide

Keep, change or drop — with evidence, and humans in charge.

Principles

Evidence over output

More screens is not the point. Every prototype exists to answer a question.

Real over representative

Real data, real devices, real scenarios — including the messy ones lorem ipsum hides.

Humans decide

AI drafts, explores and checks. People review, choose and stay accountable.

How it works in practice

Rapid prototyping with real data

Instead of static mockups, I build coded prototypes that run on realistic or anonymised production data and on the scenarios that actually happen: devices going offline, empty histories, values that don’t add up, a fleet of 200 chargers instead of three.

  • From idea to clickable, data-driven prototype in hours
  • Real edge cases surface before development, not after release
  • Stakeholders react to something that behaves like the product

Quick validation

Because prototypes are cheap, I test more often and earlier — and throw more away.

  • Several variants tested side by side with users
  • Short moderated sessions and unmoderated tests on working prototypes
  • Feedback, transcripts and reviews synthesised into themes the same day
  • Decisions documented with the evidence behind them

Agents in the workflow

Small, focused agents take over the repetitive parts of design and product work, always with a human in the loop.

  • Research agent: clusters interviews, tickets and app reviews into insights
  • Spec agent: drafts user stories, acceptance criteria and edge cases from flows
  • Design QA agent: compares builds with the design system and flags drift
  • Docs agent: keeps release notes and documentation up to date

From design system to component architecture

A design system only pays off when it matches the code. I structure tokens and components so they map one-to-one to the real component architecture.

  • Tokens (primitive → semantic → context) shared by design and code
  • Components named, structured and documented like their code counterparts
  • Prototypes built from the real components, so “done in design” is close to “done”
  • Themes and white-label brands generated from tokens, not duplicated screens

Design operations

AI also takes friction out of how the team works together.

  • Consistency and accessibility checks across files and screens
  • Faster handoff: specs, states and copy prepared for engineering
  • Workshop and review summaries with clear decisions and owners

Guardrails

Speed only matters if people can trust the result.

  • No customer or personal data outside approved tools
  • Every AI-generated artefact is reviewed before it’s shared
  • Accessibility, security and brand rules are part of every check
  • Clear about what was generated, and what was decided by people