How I Work With AIa point of view
A point of view · from practice

I don’t use AI. I build with it.

A chatbot answers questions. I build the systems that do the work — and I bring the judgment that makes them right.

I’m a product designer. Over the last year I’ve used AI to build a design system, a self-healing backup that runs while I sleep, a dozen apps, and the very site you’re reading this on. None of it was prompt-and-paste. This is how I actually work with AI — and why.

throughput × judgment
judgmenttaste · standardsthroughputmemory · speedthe gateships
AI-in-the-loopSystems & automationDesign-buildProductReverse-engineeringHuman-in-the-loop
01Why I build

The instinct came first. The tools just caught up.

The pattern is always the same. I needed something and no tool did it well, so I built it. I wanted to see an app exist, so I made a concept to find out what it would do. I build to make my life easier — and to chase the things I love.

“I needed to do something, I couldn’t find a tool to easily do it for me, so I built it. I wanted to see an app in the world, so I built a concept to see what it would do.”
Out of love

SaberIQcame from wanting to teach my nephew the subtleties of baseball — a game I love — so I built him a way to see them.

Out of need

The backup system came from needing my work to survive a bad day. Same instinct, pointed at a different problem.

02Pair, not autopilot

I don’t hand it the wheel. I pair with it.

The shift that changed everything was refusing to treat AI as a fancier autocomplete.

“I stopped treating Claude as a fancier autocomplete and started treating it as a design-engineering pair.I brought the taste, the standards, and the ‘why.’ It brought the throughput and a memory that never lost the thread.”

A pair, not an oracle. I stay the one who decides — which is exactly where the next part comes in.

03The reframe

Judgment is the scarce resource now — not production.

Here’s what most people miss: AI collapses the cost of making the artifact. It does nothing to the cost of deciding what’s correct.

“Claude was only ever as good as the standards I set. The bottleneck was never its speed. It was my judgment.

So the work moved — less production, more deciding. That’s not a downgrade. It’s the most senior thing a designer does, finally made visible.

04How I actually work

Five moves that turn a fast tool into a trustworthy one.

Aim at infrastructure, not artifacts

Anyone can get an AI to make a button. The leverage compounds when you aim it higher — at the infrastructure that keeps a whole system honest.

Specs are the leverage

A vague prompt gets a mediocre component; a sharp standard gets a system. Writing the standard is the skill.

Verify — never trust the “done”

It has told me components were fixed that weren’t, and told me things were backed up that weren’t. My eyes caught what its own scans cleared. Speed without a gate is just faster drift.

Correct the system, not the instance

When it gets something wrong, I make it encode the distinction — so the same mistake can’t come back next time.

Restraint

When AI makes iteration nearly free, the scarce thing isn’t production — it’s restraint. Knowing what not to ship is the taste.

05What it produced

Not theory — the actual work.

A technical path and a creative one, plus the design engine underneath both. Each of these is becoming its own case study — linked here as they land.

Creative — apps born of need & passion
SaberIQSabermetrics with an AI explanation on every stat — built to teach my nephew the game.
PT BuddyA PWA that auto-books my physical-therapy sessions against my calendar.
Quiet JoyA native iOS gratitude app, full managed backend and all.
And the honest tell: this very site is the artifact. These case studies live inside a portfolio my AI-build method produced. The medium is the proof.
What it comes down to

Good work shouldn’t need permission to survive.

I build with AI as infrastructure — and as a way to chase what I love. The tools got faster. The judgment is still mine.