The Vibe Father
Playbook
A Mac out of the box is about forty-five minutes away from being a serious AI coding machine. This is the walkthrough we give friends who ask — every command, in order, with the safety rails installed before the first agent runs rather than after the first mess. Nothing here requires our app or anyone else's; it's the foundation everything sits on.
Step 1: Foundations
Three things everything else depends on.
Xcode Command Line Tools. Compilers, git, and the headers half your future installs will silently need:
xcode-select --installClick through the dialog; it takes a few minutes. (The full Xcode app is not required unless you're building Mac/iOS apps.)
Homebrew. The package manager the whole Mac dev ecosystem assumes. Get the current install command from brew.sh — run it from the site, not from a blog post's possibly-stale copy — and follow the post-install instructions it prints about adding brew to your PATH.
Git identity. Agents will be making commits on your behalf soon, so tell git who you are:
git config --global user.name "Your Name"
git config --global user.email "you@example.com"
git config --global init.defaultBranch mainStep 2: Runtimes
The AI coding CLIs are distributed mostly via npm, pip, and brew, so install the runtimes once and everything downstream is a one-liner:
brew install node python git ghThat's Node (for npm-installed CLIs), Python (for pip-installed ones and half your future scripts), a current git (newer than the Apple-bundled one), and GitHub's gh CLI — which agents can use to open PRs, read issues, and check CI, so it earns its place even if you never type it yourself. Run gh auth login once and follow the prompts.
Step 3: Pick and install your coding CLI
Don't install five; install one or two and get fluent. The honest decision process is its own post — How to Choose an AI Coding CLI walks the five questions — but the short version: if you want a single lab's flagship experience, take that lab's CLI; if you want to point one tool at any model, take a provider-agnostic one like Aider or OpenCode.
The majors all install through the package managers you just set up. The one install command we'll quote is the documented one for Claude Code:
npm install -g @anthropic-ai/claude-codeFor the others — Codex CLI, Gemini CLI, Aider, OpenCode — use the install command from each project's own site or README rather than a copy here; these tools ship fast and their install paths change. Whichever you pick, run it once in an empty directory and complete its login or key setup before pointing it at real code.
Step 4: Keys, with spend caps
If your CLI uses API billing (or you want the freedom to), provision your keys properly: create the key in the lab's console, set a monthly spend cap before doing anything else, store the key in a password manager or the macOS Keychain, and never let it near a file that could reach git. We wrote the lab-by-lab walkthrough — consoles, first steps, and what an evening actually costs — in the BYOK setup guide. Ten minutes per lab; the spend cap is the step that makes every future mistake boring.
Step 5: Safety rails, before the first real run
This is the step people skip and regret. Agents edit files and run commands; your protection is git, used with intent:
- A git repo for everything. Every project an agent touches — including experiments, including "just a script" — starts with
git init. No repo means no undo, and no undo means you'll supervise every edit forever. - The checkpoint habit. Before every autonomous run:
git add -A && git commit -m "checkpoint: before X". Three seconds. When a run goes wrong,git reset --hardmakes it not have happened. Squash the checkpoints into real commits before merging. - Worktrees for parallel agents. The day you run two agents at once, give each its own directory with
git worktree add— never two agents in one working copy. The full playbook, with all the commands, is in Checkpoints & Git Worktrees.
Install these as habits now, while the stakes are zero. They're the difference between agents you can trust with real work and agents you babysit.
Step 6: Terminal quality of life
Any terminal works — Apple's built-in Terminal.app runs every tool in this guide without complaint, and there is no productivity prize for switching. That said, most people doing this daily end up in a third-party terminal for quality-of-life reasons: iTerm2 (split panes, deep configurability), Warp (modern UI, command palette), and Ghostty (fast, minimal) are the ones we see most. Try one if you're curious; skip it if you're not. Two settings worth having anywhere: a bigger scrollback buffer (agent output is long) and a font size you can actually read at the end of the day.
Step 7: The optional command deck
Everything above gives you a complete, safe, single-agent setup — and for a while, that's genuinely all you need. When you graduate to running multiple agents, multiple CLIs, and review gates across them, that's the layer we build: The Vibe Father is a Mac-native command deck that sits on top of the CLIs you just installed. One download at /download/mac if and when you get there. It's step seven for a reason — get fluent with the fundamentals first.
Step 8: The smoke test
Prove the whole stack with one small real run:
mkdir ~/code/smoke-test && cd ~/code/smoke-test
git init
git add -A && git commit -m "checkpoint: empty" --allow-emptyLaunch your CLI in that directory and give it a small, concrete task — "write a Python script that fetches a URL and prints the title, with error handling, and a test" is a good shape. Watch the loop: it should create files, run them, see failures, and fix them. When it finishes, read the diff with git diff, run the test yourself, and commit. If all of that worked, your machine is ready for real projects.
Where to go next
You now have the setup; what's missing is the mileage. Pick a small real project — something you actually want to exist — and build it end to end with your new stack. We wrote the roadmap for exactly that in Ship Your First Product with AI Agents: scoping, prompting, verifying, and getting the thing in front of people. The setup took an hour; the fluency comes from shipping.