Find the problem
before you touch anything.

Three careers, one instinct running under all of them. This page sets out what that instinct actually is, and what it means for the way I design and build interfaces — the argument a three-line CV compresses into trivia.

Three signals, one instinct

Diagnosis from incomplete signals

Seven years, veterinary surgery

Emergency, surgery, general practice — thousands of cases where the patient could not describe the problem. You read behaviour, form a hypothesis, and test it before committing to a treatment, because you live with being wrong. Reading why a page fails to convert, or why a system breaks on one input path nobody flagged, is the same exercise: observe what is actually happening, not what the brief assumes is broken. It showed up in the numbers even then: running a branch clinic single-handed I took weekly earnings 14% above the group average, and tripled the dental procedure rate against clinic norms, by fixing how the consulting schedule and the standard of care were structured rather than by working longer.

Knowing what a buyer actually responds to

Startup operator, Vietnam and Australia

Product manager for three products at a fin-tech SaaS company, where I integrated three separate development teams of eight under a LeSS framework and managed a pivot from a directory to a booking platform. Before that, co-founding a startup that went from zero to 600 partner locations in six months. Watching the same pitch close in one room and die in the next teaches you which decisions are about outcomes and which are about house style. Most weak interfaces are designed by people who never had to sit across from the buyer.

A one-person practice with machine leverage

Now, independent

Solo, running an agent fleet: Claude Code as orchestrator, subagents doing scoped work in parallel git worktrees, an always-on VPS, memory that survives between sessions, enforcement hooks that block an unverified claim, and a compounding post-mortem loop. The engine behind it: 66 repos in 18 months — experiments cheap enough to run constantly, each one feeding the systems that ship.

Three careers as stacked layers, with one instinct running through all of themVeterinary surgery, startup product and sales, and a solo practice with an agent fleet, drawn as three stacked strata. A single line labelled "find the problem before you touch anything" cores down through every layer, because each one taught the same instinct.Veterinary surgerySeven years · the patient cannot tell you what is wrongStartup product and salesThree products to an ASX listing · buyers say one thing, do anotherSolo practice, agent fleetSince 2019 · the machine will confidently build the wrong thingFind the problembefore you touch anythingRead top to bottom. Same lesson, three different ways of being taught it.Three careers as stacked layers with one instinct through all of themVeterinary surgerySeven years · the patient cannottell you what is wrongStartup product and salesBuyers say one thing andthen do anotherSolo practice, agent fleetThe machine will confidentlybuild the wrong thingFind the problem before youtouch anythingSame lesson, taught three ways.

What that looks like in the work

The instinct is only useful if it shows up in the craft. This is the deeper version of the skills list on the resume.

Design systems

A design system is an argument written down — a documented palette, type scale, motion rules, and the do’s and don’ts that keep a hundred decisions consistent without a meeting. This site runs on one, revised in the open as the work exposed where it was wrong.

Interface and motion craft

Interaction detail: hover states, transitions, motion that carries meaning instead of decoration — and every animated state ships with a reduced-motion alternative and a no-JS fallback, so nothing depends on the motion running.

Front-end engineering

Astro, React, TypeScript, and Canvas — used for what each is good at rather than forced into one stack for everything. Astro for a fast static shell, React where a page needs real interactivity, Canvas2D for anything that draws its own frames.

Agent orchestration and context engineering

An orchestrator that plans and delegates, subagents scoped to one job each, parallel git worktrees, and hooks that block a “verified” claim without an artifact behind it. This is the actual day-to-day workflow, not the pitch version.

Copywriting and narrative

Every heading, caption and case file on this site is mine, and the deck work is the same skill under a harder constraint: an investor slide has to land a claim in one line with the evidence visible beside it. Roughly 30 enterprise partner decks, and the materials behind $32M+ closed. Interface copy is the same job at smaller scale — a button label is an argument about what happens next.

Lifecycle automation

A Brisbane brokerage's lifecycle system, running continuously since 2019: a pre-approval expiry ladder, a five-part finance campaign, and a post-settlement programme that keeps talking for three years. The finding that justifies it — 14–46% of settled loans first went dormant, and the longest revival sat dormant 750 days before settling. Automation is a design problem about timing and tone, not a rules engine.

The work is the argument.

Everything above is a claim. The pieces are the evidence.