Everything You Need Before Your AI Employee Starts

Before you can have an AI employee running 24/7, you need a few things in place. None of this is hard, but getting it done beforehand means your agent can go live in hours instead of days.

Here’s everything, step by step.

1. The brain: Mac Mini M4

Your AI agent needs a computer to run on. Not your laptop — a dedicated machine that stays on 24/7, plugged into power and internet in your office or closet.

The Mac Mini M4 is perfect for this. It’s silent, uses about 5 watts at idle (less than a lightbulb), and has more than enough power to run an AI agent.

Which model: The base model (16GB RAM, 256GB) is fine for one agent. If you think you’ll run multiple agents or do heavy work, get the 24GB model.

Setup:

• Plug in power and ethernet (WiFi works but ethernet is more reliable)
• Create a local user account — don’t use your personal Apple ID
• Turn off sleep: System Settings → Energy → “Prevent automatic sleeping”
• Turn on auto-restart: System Settings → Energy → “Start up automatically after a power failure”
• Enable SSH: System Settings → General → Sharing → Remote Login

That last step is important — it’s how you (or your support person) can fix things remotely without touching the machine.

2. The AI: Claude Pro Max

Your agent uses Claude to think. It’s the difference between a chatbot and something that can actually reason about your business.

Go to claude.ai and subscribe to Claude Pro Max ($200/month). Yes, it’s the expensive one. The regular Pro plan doesn’t include enough API calls for an agent that runs around the clock.

After subscribing:

1. Go to console.anthropic.com
2. Create an API key
3. Save it somewhere safe (you’ll need it during setup)
4. Set a billing alert so you don’t get surprised

$200/month sounds like a lot until you compare it to the $4,000-6,000/month you’d pay a human to do the same work.

3. Secure remote access: Tailscale

You need a way to reach your Mac Mini from anywhere — to check on your agent, fix issues, or make changes. But you don’t want to open ports on your network or mess with VPNs.

Tailscale solves this perfectly. It creates an encrypted connection between your devices without any network configuration. Free for personal use.

On the Mac Mini:

brew install tailscale
tailscale login

Sign in with your Google/Microsoft/Apple account. Note the IP it gives you (starts with 100.).

On your laptop/phone: Install Tailscale there too. Now you can SSH into your Mac Mini from anywhere:

ssh user@100.x.x.x

No port forwarding. No static IP. No security risk. It just works.

4. An email address for your agent

Your agent needs its own email. Don’t use your personal inbox — give it a dedicated address so you can control what it sees and sends.

Easiest option: Add a Google Workspace seat ($7/month) — something like assistant@yourcompany.com. This gives the agent its own inbox, calendar, and drive.

Free option: Create a Gmail alias in your existing account. The agent reads via API and sends as the alias.

Start conservative: Give it read-only access first. Let it draft responses for you to review. Once you trust it, you can let it send on your behalf.

5. A messaging channel

This is how you’ll talk to your agent day-to-day. Like texting an assistant.

My recommendation: Telegram. It’s free, works on every device, and setting up a bot takes 2 minutes:

1. Open Telegram, search for @BotFather
2. Send /newbot
3. Name it (e.g., “My AI Assistant”)
4. Save the token it gives you

Other options: Slack (if your team is there), iMessage (if you want it to feel like texting a person), or WhatsApp.

6. A password manager

You’re going to have API keys, bot tokens, and credentials for various services. Don’t put these in a text file.

Use whatever you already have — 1Password, Bitwarden, LastPass. Create a folder specifically for your agent’s credentials.

The checklist

Before setup day:

• Mac Mini M4 ordered and plugged in
• Claude Pro Max subscribed, API key saved
• Tailscale installed on Mac Mini and your devices
• Agent email address created
• Telegram bot created (or other messaging decided)
• Password manager ready with an “AI Agent” vault

That’s it. Everything else — installing OpenClaw, configuring your agent, building the memory system, connecting your tools — happens during setup.

---

Want someone to handle all of this for you? I do on-site and remote installations — you get a working AI employee without touching a terminal.