Your first useful AI task.

Start with a read-only result in a test project. Budget about 70 minutes for the full checklist, plus account setup and troubleshooting; it is not a 30-minute requirement before AI becomes useful.

For the first session, complete the safety checks below, follow the native-install quickstart, and ask for a summary of a project you know. Compare it with the files. GitHub hosting, deployment, a vault, connectors, scheduling, and swarms can wait until a real task needs them.

Before connecting accounts or scheduling work

• Read Chapter 9 now, not at the end of the day. Use synthetic data and a test account; keep customer data and production credentials out of the first session.
• Keep permission prompts enabled. Review every connector's scopes and destination; start with read access and require human approval before sending, publishing, or changing records.
• Choose a spend limit for metered API work, name the workflow owner, and verify the failure notification before the first unattended run. A successful empty result and a failed read are different outcomes.

Deployment is optional. Vercel Hobby is for personal, non-commercial use, not every product in this book. Check the relevant plan before using it for a business. Installation and plan guidance checked on 2026-09-05.

Where projects live (folder hygiene that survives 30 days).

The first question most non-technical operators ask after installing Claude Code isn't "what does it do." It's some version of "where do I keep my projects so this isn't a mess in 30 days." Nobody covers this. Then on day 31 you have eleven half-named folders on your Desktop, three of them have a .claude/ inside, two of them are actually the same project, and you can't remember which one Claude is reading from. Decide the structure now. Five minutes here saves a Saturday in June.

Here's mine, in three lines. Code projects live at ~/Desktop/<project-or-company>/ — one folder per repo, folder name matches the GitHub repo name matches the .claude/ scope. The Obsidian vault lives at ~/Desktop/Obsidian/<vault-name>/ — separate top-level home, never inside a code repo. Nothing else lives on the Desktop. No "AI Stuff." No "experiments." No "temp." If it's a project, it gets its own folder at the right level. If it's not a project yet, it sits in your vault inbox until it earns one.

The rule is one folder, one project, one CLAUDE.md. The CLAUDE.md at the repo root is the handbook every new session reads on wake-up — see Ch 4 — the vault for the same pattern at the vault level, and the day-zero step "Drop a CLAUDE.md at your repo root" below for the file itself. Per-project context stays inside that folder. Personal context (you, your accounts, your tone) stays at ~/.claude/CLAUDE.md. They compose. They don't collide.

Two anti-patterns that bite operators about three weeks in. First: putting your Obsidian vault inside a code repo. It sounds tidy. It ruins both — git starts tracking thousands of markdown files, Obsidian sync fights with git, and Claude reads notes that belong to a different mental scope when you're trying to ship code. Second: the "AI Stuff" mega-folder. One folder, six projects, four abandoned, two active, all sharing one CLAUDE.md that contradicts itself. The agent reads context that's 70% irrelevant and you wonder why the outputs feel scrambled. Don't do this. One project per folder. Always.

If your machine is already messy: don't reorganize the world today. Pick the one or two projects you're actually working on this week, mv them into the structure above, drop a CLAUDE.md at the root of each, and leave the rest of the mess where it is. The unused folders aren't hurting you — they're just clutter. You can sweep them on a Sunday. Day zero is for the projects that matter this week.