Pair-designing with a stubborn senior

This site was built with a design tool that opens by interviewing you. Not prompting. Interviewing. It refuses to write a single line of code until it has dragged a strategy out of you that fits on one page.

Round one was audience and outcome. Who lands here, what should they leave with. Round two was personality and references. Three words for the brand voice. Anti-references: which five common shapes this site must not become. The artifact at the end was strategic, not visual. Register, voice, principles. No colors. No fonts. No widths. Those came later, and they came easy because the strategy already constrained them.

I asked for five visual directions before committing. The tool resisted, then complied: terminal CRT, brutalist slab, holographic dual-tone, neon vector grid, glitch magazine. Same content, five skins, switchable from a corner button. I picked the holographic one. The other four got deleted the same hour. Optionality is expensive; keeping it past the decision point is more expensive.

The discipline I did not expect was negative. The tool maintains a list of patterns that read as AI-generated on sight: side-stripe borders, gradient text, glassmorphism as default, the metric-grid hero, identical card decks, modals as the first answer to any question. When I asked for a gradient title, it refused the gradient and rewrote the effect with offset shadows. Same look, none of the tell. That refusal is the actual deliverable.

The other surprise was that visual register and copy register are independent. The holographic look pulled the writing toward radio chatter. TRANSMISSION OPEN, SIGNAL +12dB, frequency labels everywhere. I cut all of it late in the session. The visuals stayed; the copy got plain. Loud framing, dense interior. Two dials, turn them separately.

What I am taking from this beyond the site: write the strategy down before the visuals. Keep an explicit list of the shapes you refuse to become. Treat refusal as a feature. The work it produced will keep proving why.