How this was built
My own framework, applied to my own website.
I talk a lot about Machine Work vs. Meaning Work — let AI absorb the repetitive so humans can do the judgment. This site is that idea, executed. Here's exactly how this was built.
Two hours, PRD to production
I wrote a product requirements document the way I would for any enterprise build: goals, non-goals, content architecture, acceptance criteria. Then I handed it to an AI coding agent — OpenAI's Codex, running on my Windows machine — and the first version of this site went from empty repo to production in about two hours. It has kept evolving since, the way real products do.
The agent did more than write code
The domain's DNS still pointed at a system I retired years ago, and email routing was broken. Using computer use, the agent reconfigured the DNS records and fixed the email setup while I supervised. Machine Work, handled.
Where the human stayed in the loop
Every factual claim, every privacy decision, every word of voice on this site is mine. The AI drafted, structured, and implemented; I decided what's true, what's public, and what sounds like me. That division of labor isn't a limitation of the tools — it's the correct governance model, and it's the same one I bring to enterprise AI adoption.
The architecture
Career history, resume variants, and proof points live as structured data, separate from presentation. Astro renders static pages with almost no client-side JavaScript; Tailwind keeps the design system restrained; Vercel handles deployment. Adding a tailored resume variant means editing data, not rebuilding pages. No chatbot, no tracking stack, no CMS — deliberately. Fast and credible beats clever.
If you want to talk about what this looks like at enterprise scale — with compliance, PHI, and audit trails in the room — that's my day job.