Your first build

The goal of this track is simple and non-negotiable: get something live. Not a tutorial you followed along with — a URL you can send to a friend. Everything you learned in Track 0 (the loop, directing instead of typing, git as your undo button) you now put to work.

The drill: same project, any tool

You’ll build a personal landing page and deploy it. It’s deliberately small, because the point isn’t the project — it’s learning to run the loop cleanly:

1. Give a plan, not just a wish. “Build a one-page site with a hero, three sections, and a contact button, using plain HTML and CSS” beats “make me a website.”
2. Let it work, then read the diff. Don’t rubber-stamp. Skim what changed and why. This is the habit that separates people who ship from people who get stuck.
3. Run it. Let the agent see the result. Preview the page. If it’s broken, paste the error or describe what’s wrong — feed the loop.
4. Ship it. Deploy to a free host (Cloudflare Pages, Vercel, or Netlify). Done means live.

Do this once in your tool of choice, then — if you have the appetite — do it again in a second tool. You’ll notice the commands differ but the rhythm is identical. That’s the transferable skill.

Each tool’s dialect (the 10% that differs)

• Claude Code — terminal-first. You’ll write a short CLAUDE.md so it knows the project, and learn permission modes so you’re not approving every step.
• Cursor — if you want a familiar VS Code window. Learn @-referencing to give it context and Agent mode for multi-file edits.
• Codex — terminal-first like Claude Code. Learn its three approval modes and the AGENTS.md context file.
• opencode — bring-your-own-model. Great for feeling the model-vs-harness idea from Track 0 first-hand.

Projects in this track

Start with the landing page, then climb: a CLI habit tracker (let the agent run and test its own code), an “explain this repo” exercise (context referencing, not generation), and a small Chrome extension (working inside a platform’s rules). Each one is done when it’s real — see the project ladder for the exact “done” criteria.

What “good” looks like

• You shipped at least one project to a public URL.
• You reviewed diffs instead of blindly accepting them.
• You recovered from at least one mess with git instead of panicking.
• You can describe how your tool’s loop felt — where it flew, where it stalled.

Next up, Track 2: the context and workflow habits that make the agent dramatically better — CLAUDE.md/AGENTS.md, working from a spec, and building your own commands.

This track is being written in full now — the tool-by-tool walkthroughs land with launch. Join the newsletter to get them first.