This website

Directing instead of making

I didn't write the markup for this page. I described what I wanted, reacted to what came back, and kept redirecting until it was right. That's closer to how I spend my actual workday than any side project I've built.

It turned out to be a good test of a question I care about at work: where do AI tools genuinely help design, and where does human judgment still have to carry it?

The clay family, and the walk that didn't work

I wanted my family on the homepage: the four of us and the dog, in clay, out on a mountain. The first idea was a walk cycle: they'd stroll in from the left and sit down at the base.

It never read right. It looked like shuffling, then jigging, then walking on an escalator. Each fix surfaced a new problem: frames that were near-duplicates, the same leg leading the whole cycle, travel speed that didn't match the stride.

Then I cut most of it. What's there now is a short walk-in, followed by seven frames that settle, crouch, turn forward, and form a seated group.

Taste is the bottleneck, not output

Producing variations became nearly free. Deciding between them did not.

Vague reactions were still useful, because a good partner can measure its way toward the cause.

Accessibility didn't come for free

The palette I liked had muted greys that failed contrast at small sizes. The homepage's own navigation measured 3.5:1 against a 4.5:1 requirement. It looked fine. It wasn't.

I kept the brand accent at heading scale, where it passes, and introduced a darker variant for small text.