Building in Public, Day One ~/jedson

// how this started

This website didn't exist this morning. By the end of the day it was live, deployed, and sitting at its own domain. That's not a brag — it's kind of the point. I wanted to see what it actually looked like to build something in public with AI as a genuine collaborator, not just a code autocomplete.

The starting point was a template — a dark terminal-aesthetic HTML file with placeholder content. Sections for projects, recipes, and a resume. A design I liked the bones of but that had exactly zero of my actual information in it. ~/yourname. Based in — —. Recipe One. The whole thing was a skin waiting to be filled.

// the process

I opened a conversation with Claude and pasted in the HTML. Then we just... talked through it. My name first. Then the bio — I rambled, Claude distilled. I told it I'd been hacking since accidentally finding a BASIC prompt on an Apple IIe back in 1985, that I've been professional since 2011, that I'm trying to pivot toward consulting. It turned that into a two-sentence intro that actually sounds like me, which is harder than it sounds.

The cooking section came next. I mentioned I'm a beginner in the kitchen and that my wife is the one who decides what's actually worth sharing. Claude added a line about her being "the official quality filter" — I kept it. It's accurate.

We added the Upwork badge, filled in my location (Mendocino County, CA), dropped in GitHub and Bluesky icons in the nav with SVG logos tinted in the site's green. There were a couple of wrong usernames, a couple of broken links. Iteration. Nothing dramatic.

On the Trading Reflex Simulator entry: I'd already built that app — an experiment in multi-agent AI development using coordinated AI specialists alongside human judgment. Got it from concept to a working iOS/Android app with in-app purchases in under two weeks. Writing it up for the site forced me to articulate what I actually learned from it: AI agents are excellent at grunt work, asset discovery, and procedural generation. Humans are still essential for product vision, aesthetic judgment, and feel.

// where it landed

By the end of the session: a live personal site with two real projects, one real recipe (Broccoli Butter Fettuccine — started as an attempt to recreate Pasta-Roni from scratch, turned out better), a work background section, social links, and a disabled resume button sitting in yellow at 40% opacity waiting for me to actually write the resume.

The whole thing is a single index.html and one recipe page. No framework, no build step, no dependencies. Deployed on Vercel. DNS propagated the same afternoon.

// the handoff

At the end of the claude.ai session I generated a PROJECT.md — a context document designed to give a fresh AI enough background to pick up without re-litigating every design decision. Then I opened Claude Code in the terminal, pointed it at the repo, and asked it to familiarize itself with the file.

The first thing it did was restructure the projects section. The old layout was a static card grid — two project cards sitting there, one of them with a description so long it was clearly straining the format. Claude Code proposed replacing it with a post feed: dated entries, flat list, no intermediate index pages. That's what you're reading inside right now. The card for this website linked nowhere; this entry links here.

Note to self: Writing "we'll hand this off to Claude Code" as a future step, and then having the first thing Claude Code does be restructure the section that contains that sentence, is exactly the kind of thing this site is supposed to document.

I've been thinking a lot about what it means to build with AI rather than just using it. This site is one answer to that question — small, honest, documented. We'll see where it goes.