Last month I ran the numbers on what it would cost to hire a part-time executive assistant. Even at 20 hours per week and $25/hour, that is $2,000 a month before taxes, benefits, or the inevitable 3-week training period where you are spending more time teaching them than saving.

I run five businesses. I need someone who can triage my inbox, check my calendar, process my scattered brain dumps at 2 AM, monitor financials across every entity, and — this is the important part — do it without me being the bottleneck.

So I built one. With AI. For $20 a month.

This is not a theoretical experiment. This system has been running every day for the past 30 days. It processes tasks while I sleep. It sends me a morning briefing before my alarm goes off. It triages new ideas into six categories before I have had my first cup of coffee.

Here is exactly how it works.

The Problem: You Are Not the CEO — You Are the Router

If you run even two businesses, you know the feeling. You are not making strategic decisions most of the day. You are routing information. This invoice goes here. That email needs a reply. This task should have been done last Tuesday.

You are a $200/hour brain doing $15/hour work.

The standard advice is "hire a VA." I have tried that. Here is the issue: a VA still needs you to be the router. They ping you in Slack, ask for context, wait for your reply, then execute. The bottleneck just moved — it did not disappear.

What I needed was someone who could look at the full picture across all five businesses and make the routing decisions themselves. Only pulling me in when something actually needed my judgment.

That is what an AI Chief of Staff does.

The Architecture: Hub-and-Spoke

The system has three layers:

Layer 1: The Chief of Staff (the hub)

One AI agent that runs persistently. It checks email, calendar, task queues, and financials on a schedule — every two hours during waking hours, every four hours overnight. It triages everything it finds using the same framework I would:

Layer 2: Specialized agents (the spokes)

Each business gets its own AI "CEO" agent. These agents have narrow scope — they only care about their one business. They check for tasks, execute within their permissions, and report back to the Chief of Staff.

No agent talks to another directly. Everything routes through the hub. This prevents chaos and ensures nothing falls through the cracks.

Layer 3: Me (the approver)

I wake up. I read a briefing that has already been compiled. I approve, reject, or redirect items. That takes 10-15 minutes. Then I do the work only I can do — the strategic thinking, the relationship building, the creative work.

Everything else is already handled.

The Stack: What It Actually Costs

Here is the infrastructure, with real costs:

ComponentPurposeMonthly Cost
Mac Mini (M4)Always-on server for agents$0 (one-time purchase)
NAS (2-bay)File storage, Docker containers$0 (one-time purchase)
Claude APIAI backbone$20
SupabaseTask queue, database$0 (free tier)
TelegramHuman interface$0
DockerContainer orchestration$0

Total recurring cost: ~$20/month.

Compare that to a part-time EA at $2,000/month. Or a full-time Chief of Staff at $8,000-$15,000/month. Even a premium virtual assistant service runs $1,500-$3,000/month.

The hardware was a one-time investment. A Mac Mini M4 runs about $500. A basic NAS with two drives is another $300-$400. Under $1,000 upfront, $20/month forever after. That is the price of a Netflix subscription and a coffee.

What It Actually Does Every Day

Here is a real day from last week:

6:45 AM — Morning briefing arrives via Telegram:

I spent 12 minutes reviewing. Approved the pricing decision. Dictated two email replies (drafted by the AI, I just tweaked them). Done.

While I slept the night before:

None of that required me to be awake, online, or thinking about it.

The Brain Dump System: Capture Everything, Triage Later

The most underrated part of this system is the brain dump workflow. Here is how it works:

  1. I have a thought — at any time, any place
  2. I text it to my Chief of Staff via Telegram (voice or text)
  3. The AI captures it and categorizes it: DO / DELEGATE / SCHEDULE / DEFER / REFERENCE / TRASH
  4. It routes delegates to the appropriate business agent
  5. It confirms the routing with me

This replaced Notion, Todoist, Apple Reminders, sticky notes, and the "send yourself an email" hack. Everything goes to one inbox. AI triages it. I just review.

The key insight: your brain is for generating ideas, not storing them. The faster you can dump a thought into a trusted system, the faster your brain lets it go and moves to the next one.

Why This Beats Hiring (For Now)

I am not anti-hiring. When the time is right, I will bring on people. But here is what an AI Chief of Staff does that a human EA often cannot:

  1. Works 18 hours a day without overtime, complaints, or burnout
  2. Scales across all businesses — a human EA usually specializes in one
  3. Gets smarter every day — new instructions stick permanently, no retraining
  4. No management overhead — it does not need PTO, performance reviews, or motivation
  5. Costs less than a dinner out — $20/month is essentially free for a business

The trade-off: it cannot make judgment calls in ambiguous situations the way a great human can. That is why I am the approver, not the AI. It handles volume and routing. I handle strategy and relationships.

How to Build Your Own

If you are running multiple businesses or side hustles and you are drowning in context-switching, here is the roadmap:

Week 1: Set up the brain

Week 2: Build the spokes

Week 3: Connect your inputs

Week 4: Tune and trust

The full system, with templates, identity documents, and step-by-step setup instructions, is in The Axe Playbook.

The Bottom Line

Running multiple businesses does not have to mean running yourself into the ground. The technology exists today to build yourself a tireless, focused, reliable Chief of Staff for the cost of a streaming subscription.

The question is not whether AI can help manage your businesses. It is whether you are willing to invest four weekends building the system that lets it.

I invested four weekends. Now I spend 15 minutes a day in "approve mode" while the system handles the rest.

That is the trade I would make every time.