Day One · Session 03 · Dates TBD
A marketing site, built and deployed in one sitting
Ship AIPhoenix, ArizonaPress → to begin
Everything so far has lived in a chat window. Tonight it gets a URL.
Same business, same files, same voice. A stranger can finally read it.
Four sessions, one business — yours. Every session is free and standalone, each one leaves a folder the next one reads, and catching up late is explicitly allowed.
Act one
Claude Code and Codex are the same model you've been talking to for two sessions. The difference is that they can open your files, run the build, and read the error themselves.
The chat window
The agent in your repo
It's a terminal, which is the part that puts people off. Nothing tonight needs you to memorise a single command.
Setup, together
npm i -g @anthropic-ai/claude-code && claude
node v22 — oklogged inready in ~/day-one
The slowest fifteen minutes in the program, and we spend them as a room: Node, an account, an agent that starts, the template cloned. Nobody moves on while the person next to them is still installing.
Instructions went out in Discord before tonight, and this session needs a paid plan or an API key — said plainly rather than discovered halfway through. If you did the setup at home, this is your coffee break.
Act two
Every section has a job. Anything without one gets cut, however nice it looks — and the hero is a test of session 01, not a writing exercise.
If the positioning doesn't survive contact with the page, the positioning was the problem — not the page.
Scaffold, then say something
/site-scaffold → /landing-copy → /og-image
scaffolding app/ …hero ← 01-voice/positioning.mdpreview: your-site.vercel.app
Routing, metadata and sections first, because an empty page you can deploy beats a perfect page you can't. Then the copy, pulled from the one-pager you wrote in session 01. Then links that preview properly, which is ten minutes you spend once.
Proof over adjectives, all the way down. "Blazing fast" is discounted by every stranger who reads it; a number with a method behind it isn't.
Act three
One table of real data and one template becomes a page per row. Done carelessly it's spam. Done once, carefully, it's the cheapest traffic you'll ever build.
The same rows, refracted into whichever shape someone actually searches for. Your ICP file already says what they're looking for — this is that list, turned into pages that answer one query each.
The test isn't "would this rank". It's "would a person be glad they landed here".
Tonight's run
/content-map → /programmatic-pages → deploy
18 real queries → 12 pagesgenerating 5 as a sample…deployed — live in 34s
Work backwards from what your ICP types into a search box, generate five pages from real data, read one of them properly, then push. The articles you drafted in session 01 get a home on the same deploy.
Your site
Live
Not staged, not on your laptop. A URL you can paste into a message and watch someone open.
If it ships, it speaks.
Next
The site you just shipped grows a support agent that answers from your own pages, and a workflow that drafts next week's article while you sleep. Same repo, prepared branch, nothing built from scratch.
shipai.club · Free, always