Writing

Thinkingout loud.

Design systems, product decisions, and the things I've learnt building software for the past 17 years.

How we designed an AI-native notification

How does an AI-driven notification work? We created a card that quickly earns trust by analysing email tone for legal fee earners. Here's how we designed it, what we eliminated, and the platform changes we implemented.

May 5, 2026

•

7 min read

ALL POSTS

DESIGN SYSTEMS

CRAFT

PROCESS

AI + DESIGN

CAREER

BRAZILIAN JIU-JITSU

8 Articles

AI + Design

The Niland Test – Testing AI prompts at scale

How do you know a prompt is good enough to ship? Not by skimming an output. By testing it. The evaluation method I built to score prompt performance with users, turn judgment into evidence, and bring repeatable validation into AI product development.

How an AI summary becomes an action

AI features are usually judged by what they produce. The harder design problem is what happens next. The decision tree I built behind Proclaim's summarisation feature scores every summary against five criteria, picks the right three onwards actions, and turns a one-shot output into part of the working day.

How we designed an AI-native notification

Starting jiu-jitsu at 40

The year I was turning 40, I walked into a Brazilian jiu-jitsu gym for the first time. I'd talked myself out of it for years – it was too late, too expensive, and too selfish. The story of one taster class, the voice at the door telling me to leave, and why I signed up before I'd even caught my breath.

How we designed the prompts behind a Legal AI Copilot.

Prompt quality is rarely about model sophistication; it’s usually about instruction design. A practical framework of six elements I developed while building Copilot AI to help design prompts that are clearer, safer, and more likely to perform in real use.

April 29, 2026

•

6 min read

Let's talk

BLOG CV LINKEDIN

← BACK