Getting started

Start here

Everything you need to go from zero to your first working AI loop. Follow the steps in order. Should take about thirty minutes.

Key idea

The entire system is one loop: define what you want, let AI build it, then prove the result with a clean build and git proof. Every page on this site teaches a piece of that loop.

Step 1

Tools you need

Get these four things set up first. Everything else builds on this foundation.

VS Code + GitHub Copilot Tool

Download VS Code, then install the GitHub Copilot and Copilot Chat extensions. Copilot subscription is $20/month and makes writing code 3x faster.

OpenAI ChatGPT Plus

Sign up at openai.com. Plus subscription is $20/month and gives you GPT-4 access plus Operator Mode. You'll use ChatGPT to write task specs and Claude to execute them.

Terminal bootstrap

Copy and paste these commands into your Terminal to install Homebrew, Git, Node.js, and the GitHub CLI. Replace "Your Name" and email with your own.

macOS Terminal

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
brew install git
brew install node
brew install gh
git config --global user.name "Your Name"
git config --global user.email "your.email@example.com"
node --version
npm --version
python3 --version
pip3 --version
gh auth login

Note: The gh auth login command is optional. Skip it if you don't plan to use GitHub CLI.

Verify Python

Most Macs come with Python 3 pre-installed. Run python3 --version to check. If you need to install it, use brew install python3.

Reference

Detailed Setup Instructions

Step-by-step installation guides for each tool if you prefer more detail or run into issues.

VS Code

Download from code.visualstudio.com

·Required Extensions: GitHub Copilot, GitHub Copilot Chat

·Recommended: Copilot Individual subscription ($20/month)

·Why: Inline completions and chat make writing and editing code 3x faster

AI Accounts

OpenAI account from openai.com

·Recommended: ChatGPT Plus ($20/month) for GPT-4 access and Operator Mode

·Why: You will use ChatGPT to write task specs and Claude to execute them

·Optional: Claude Pro for longer context windows on complex tasks

Homebrew

Mac package manager for installing command line tools

Install

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

Verify

brew --version

Git

Version control system for tracking code changes

Check if installed

git --version

Install via Homebrew

brew install git

Configure

git config --global user.name "Your Name"
git config --global user.email "your.email@example.com"

Node.js

JavaScript runtime for Next.js and modern web development

Install via Homebrew

brew install node

Verify

node --version
npm --version

Note: Node 18+ LTS recommended for Next.js 16

Python

Most Macs come with Python 3 pre-installed

Check version

python3 --version

Install if needed

brew install python3

Verify pip

pip3 --version

Optional: Install virtualenv with pip3 install virtualenv

GitHub CLI (Optional)

Manage GitHub repos from the command line

Install

brew install gh

Authenticate

gh auth login

Terminal Bootstrap Checklist

Run these commands in order to set up your environment in one go

/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
brew install git
brew install node
brew install gh
git config --global user.name "Your Name"
git config --global user.email "your.email@example.com"
node --version
npm --version
python3 --version
pip3 --version
gh auth login

After running these, replace the name and email placeholders with your actual details.

Optional Advanced Stack

Tools you might need later for specific workflows

·Docker: brew install --cask docker

·pnpm: npm install -g pnpm

·OpenAI CLI: pip3 install openai then set OPENAI_API_KEY

Checklist

Build with me

Follow this exact path. Each step uses the operator loop: you define the goal, the AI does the work, and you verify the output before moving on. By the end you will have shipped a real change with proof.

Docs Kit sets up your repo docs. Activator runs tasks with proof.

1
Read the operator loop on Activator
Understand the six-step cycle before you run it.
2
Generate your docs kit
Scroll down on this page and run the docs generator prompt in ChatGPT.
3
Paste the Operator Mode activator into ChatGPT
This turns ChatGPT into a spec writer that produces task scripts.
4
Run your first Claude task
Paste the script Claude gives you, get a handoff packet with proofs.
5
Start Modules in order
Work through Modules 1 to 6. Each one builds on the last.
6
Use the Dashboard after membership
Track your progress once you are a founding member.

How it works

The core loop

Every task follows this cycle: plan, build, prove

The full operator loop

Six steps, every time. You stay in control at every step.

Sequence

Do this in order

There are a lot of pages on this site. Ignore most of them. Follow these five steps, in this order, and you will not get lost.

1
Start here
You are on this page. Read the first 30 minutes walkthrough below.
2
Activator
Set up Operator Mode and get your first script running.
3
Docs Kit
Templates, generators, and checklists that scaffold your project docs.
4
Modules 1 through 6
Work through them in order. Each one builds on the last.
5
Dashboard
Available after you become a member. Track your progress.

Walkthrough

Your first 30 minutes

A step by step path from nothing to a shipped, verified change. Do these in order, in one sitting.

Plan in ChatGPT. Build with Claude. Prove with checks.

Minute 0 to 5Generate your docs kitTool

Go to the Docs Kit and copy the docs generator prompt. Paste it into ChatGPT. Answer the nine questions it asks about your project. ChatGPT will output a set of markdown files.

What to paste: the docs generator prompt from the Docs Kit

Where: ChatGPT (any model)

Expected output: 7 markdown files (PROJECT, STACK, ARCHITECTURE, AGENTS, VERIFICATION, HANDOFF, UPDATE_CHECKLIST)

Create a docs/ folder in your repo root, save each file there, and commit them.

Minute 5 to 10Activate Operator Mode

Open a new ChatGPT conversation. Paste the Operator Mode script from the Activator Scripts page. This tells the AI how you work: boundaries, proof requirements, and the discipline you expect on every task.

What to paste: Script 1 (Operator Mode activation) from the Activator

Where: ChatGPT, at the start of a new conversation

Expected output: ChatGPT confirms it will follow the rules and asks what you want to build

Minute 10 to 20Run your first Claude task

Take the Claude task template from the Activator Scripts page. Fill in your repo path, branch, one file target, and a small change (add a heading, fix a typo, create a placeholder page). Paste it into Claude. Let it make the edit, run the verification commands, and produce its handoff packet.

What to paste: Script 2 (Claude task template) filled in with your details

Where: Claude (via VS Code extension or claude.ai)

Expected output: a code change, passing typecheck and build, a commit, and a HANDOFF PACKET

Pick something small for your first task. The goal is to complete the loop, not to build something complex.

Minute 20 to 30Verify, commit, and write your first handoffProof

Check what Claude produced. Run the verification commands yourself to confirm they pass. Review the diff. If it looks right, push to main. Save the handoff packet somewhere you can find it (a notes file, a GitHub issue, or your docs/ folder). That packet is your proof that the loop works.

Commands to run yourself: npx tsc --noEmit, npm run build (or your stack equivalent)

Then: git push origin main

Expected output: clean build, commit on main, a saved handoff packet with git proof

That is the whole loop. Every task you do from here follows the same cycle: plan in ChatGPT, execute in Claude, verify and ship. The rest of this page fills in the details.

Key idea

One task, one commit, one proof. If you try to bundle multiple changes, you lose the ability to trace what broke.

Want execution help?

The Operator Assistant, Activator scripts, and prompt library are available to active members.

View membership options

Stuck? Ask the Operator Assistant

Paste an error log, a failed build output, or a question about which step to take next.
Get back a bounded task script, a debug checklist, or a concrete recommendation.

Open the assistant

First project suggestion

Start with something small. A single-page app, a landing page, or one new feature in a repo you already have. The point of your first project is not to build something impressive. It is to run the full loop (spec, build, verify, ship) enough times that it becomes automatic. A todo list, a personal dashboard, or a placeholder site for an idea you have been thinking about. Pick something you can finish in a few sessions, not something that needs months of planning.

What to do when it goes wrong

Things will break. Here are the three most common problems and what to do about each one.

Claude edits files outside the hotspot

Run git diff --name-only to see what changed. If files outside your hotspot list were modified, run git checkout on those files to revert them. Then re-run the task with a stricter prompt that lists the exact files allowed. Add "Do not touch any other files" as a standalone line in your prompt.

Build fails after the change

Do not try to fix it manually. Copy the full error output and paste it back to Claude in the same conversation. Say "the build failed with this error, fix it without changing any other files." If the conversation has drifted too far, start a new one with the error and the original task. Never commit a broken build.

Handoff packet is missing proofs

If the handoff does not include git show --stat HEAD and git show --name-only HEAD, the task is not done. Ask Claude to run those commands and return the output. If it cannot (because the session ended), run them yourself and paste the results into your notes. A handoff without proof is just a claim.

Invariant

Boundaries matter

Declare which files an agent may edit. Everything else is off-limits.

What you need

All free. Nothing unusual.

·VS Code for editing and running commands
·Node.js (LTS) to run Next.js locally
·Git, starting from your first commit
·A GitHub repo as your single source of truth

The workflow

Every task follows the same basic cycle.

Phase 1

Plan

Define what you want, the constraints, and what done looks like. Do this in ChatGPT.

Phase 2

Build

Hand the spec to Claude with explicit file targets. One issue, one outcome.

Phase 3

Prove

Typecheck, lint, build. Only commit when the proof is clean.

Rule

How to talk to AI

Small habits that make a real difference in what you get back.

·One task per conversation. Do not mix unrelated changes.
·Name the file targets up front so edits stay where you want them.
·Say what done looks like so you can check the result
·Ask for build logs, type-check output, or screenshots as proof
·If things start drifting, start a new conversation instead of patching

Recommended cadence

Showing up regularly matters more than marathon sessions.

·30 to 60 minutes a day, focused on one module
·Ship one small thing each session. A route, a component, a fix
·Glance at yesterday's proof log before you start today
·End each session with a clean commit and a note on what is next

Proof

Common pitfalls

Things that tend to go wrong early on.

Anti-pattern

The most common mistake is skipping verification. If you did not run the build, you do not know if it works. Feelings are not proof.

Scope creep

One issue, one PR. If something else comes to mind, write it down and do it later.

No proofs

If you did not typecheck and build, it is not done. Always prove your changes.

Platform lock-in

Build locally, deploy anywhere. Avoid tools you cannot leave.

Mega-prompts

Long prompts with multiple asks produce worse results. Keep them short and focused.

Skipping the spec

Going straight to code means rework. Writing a quick spec first saves time.

Anti-pattern

The biggest early mistake is skipping verification. If the build did not pass, the work is not done. Always prove it.

Activator Scripts

Founding members get the full Operator Mode activation + Claude task templates. Production-ready scripts with built-in guardrails that define, execute, and verify in a repeatable loop.

View Activator Scripts See pricing

Already a member? Sign in

Ready to go?

View the curriculum Browse the Docs Kit Try the activator

Next Step

Activate the operator engine

You've learned the system. Now run it.

Define. Spec. Execute. Handoff. Verify. Ship. Repeat.

What you get with the Activator

✓7 battle-tested operator scripts
✓Structured Claude task template
✓Escalation + debug scripts
✓Cost routing + verification loop
✓Project doc generator

Unlock Activator View membership

Member resources

If you want help while you build

These tools are available to active members.

+
Operator Assistant
Ask questions about your build, debug errors, or get a second opinion on architecture decisions.
+
Activator Scripts
Production task scripts with built-in guardrails. Define, execute, and verify in a repeatable loop.
+
Prompt Library
Copy-paste prompts for specs, execution, debugging, and handoffs. Scoped and tested.

View membership options