Working with AI, with checks

The principle

AI multiplies options; it doesn't replace judgement. I use it where the work is voluminous and verifiable: synthesising research data, generating design variations to react to, and building prototypes I couldn't code alone. Decisions stay human, and checking is part of the workflow.

Three examples

1. Research synthesis

Transcripts and survey data were processed with AI, then checked against the sources so that every finding traces back to a session.

An AI-generated feedback report slide titled What's not working and why, pairing each problem with a proposed solution. — 01 An AI-generated report drawn from several interview and usability transcripts.

2. Design exploration

Layout variants, iconography and mood directions were generated quickly with Figma Make and Claude, then narrowed by hand. AI widens the funnel; judgement narrows it.

Dark and light colour schemes of the session set-up panel, shown side by side. — 02 Light and dark colour schemes explored in Figma Make.

3. Interactive prototyping

Prototypes built with Claude that would previously have needed engineering time, with hand-over notes for each branch and recommendations for building the components in Unity (C#) and React.

An interactive prototype of the character set-up screen: a humanoid rig mapping beside a 3D astronaut character. — 03 An interactive prototype in the browser, ready for early usability and accessibility testing.

In closing

AI only helps if you keep applying judgement and a narrow focus; a wide choice of options can paralyse. It is a tool, not a toy, and works best when the question is tightly defined and the result is tested.