My AI Chief of Staff: How a Team of AI Helpers Gave Me My Evenings Back

I have ideas every few minutes. A new tool for landlords. A better page for one of my apps. A fix for a property that keeps sitting empty. For years the ideas were never the problem. The hours were.

And I don't want to give up the other half of my life to get them done. I want to be in the garden with my kids. I want to travel. I want to be there for the small stuff.

For a long time it felt like I had to pick: visionary entrepreneur, or present mom. This year I stopped picking. I built an AI chief of staff named Matt. Then Matt hired his own team of 18 AI helpers. Now I talk an idea into my phone, and the team gets it done while I'm playing soccer with my kids at 6 PM.

Here is how it works, what it costs, and a step-by-step guide to build your own.

What changed, in one picture

Before: an idea went into my notes app. Maybe I got to it in a month. Maybe never. Anything that needed research, writing, and a few emails ate an afternoon I didn't have.

After: I open my phone and say what I want, the same way I'd text a friend. Something like, "Our insurance renews soon. Get me quotes from local agents and line them up side by side." Then I close the phone and go outside.

Later a card shows up on one private page. The work is done. The quotes are compared. The emails are written. There is one button: Approve.

In the first five days, from Sunday to Thursday, the team took on 101 tasks and finished 65. They logged about 1,500 minutes of work I would have done myself. That's roughly 25 hours, in five days.

AI used to chat. Now it executes.

This is the part that still surprises me. The team operates like a human team. Except the whole team costs $100 a month, as part of my Max plan on Claude Code. And they work 24/7.

The best part is zero training. I don't record videos. I don't write manuals. I simply say what I want, and they do it.

In the past, AI was a helpful chat. You asked a question, you got an answer, and then you did the work. Now it executes. It writes the email, builds the page, tests it on a phone screen, and waits for my OK.

Meet Matt, my AI chief of staff

Matt is not a person. He is Claude, Anthropic's AI, running inside a tool called Claude Code. Claude Code is the version of Claude that can actually do things, not just chat. It can read and write files, run checks, build web pages and use the apps I connect to it.

I built Matt. I wrote down who I am, what my businesses are, and how I want things done. Then I gave him the job a human chief of staff would have. He reads everything: my email, my calendar, my files, my notes, and the list of open work. He decides who on the team should do what, checks their work, and only brings me what's finished.

He checks in every hour during my workday. Each time, he:

• Reads every new idea or note I left
• Hands each one to the right helper
• Sends anything not good enough back for another round
• Writes me a short daily brief: what got done, what's moving, and the one decision that needs me

The rule I love most is the one I gave him on day one. "Don't create work for me." If the team can find a fact, they find it. If they can finish a step, they finish it. My job is the one tap only I can make.

The team Matt hired

I didn't hire these helpers. Matt did. He looked at the work coming in, decided which jobs he needed, and wrote a one-page job description for each one.

A "subagent" sounds technical. It's just a helper with one job and a one-page job description. Here's the team Matt built:

Two of these matter more than they look. Vera and Wren are the checkers. Vera opens every page the team builds, on a laptop screen and a phone screen, and looks for broken links and missing pictures. Wren reads every card against a list of the notes I've left before. If a card would earn a note from me, it goes back to its helper first.

Two real examples from the first week

1. Insurance on three rentals: 5 minutes of my time, $980 saved

Before: I'd spend an afternoon finding local agents, sending the same email five times, and building a spreadsheet as quotes trickled in.

After: Paige found local independent agents, wrote the quote requests, and lined up three carriers side by side as the replies came in. The winner kept the same deductible. My part was about 5 minutes: approving the requests and reading the final side-by-side. This requote saved $980.

2. A thin web page made deep

Before: I knew one page on an app's website was too thin. Fixing it meant a writer, a developer, and a week of back and forth.

After: Nico rewrote it to about 2,800 words with worked math, a comparison table and six new questions and answers. Vera tested it on desktop and phone. I looked at the screenshots and tapped Approve. That one was about 90 minutes saved.

What it costs

You need a paid Claude plan. Here are the prices on Anthropic's pricing page as of November 2026:

• Pro: $20 a month, or $17 a month if you pay yearly. Claude Code is included.
• Max: $100 a month, for much heavier use. This is the plan I use.
• Ultra: $200 a month, for the heaviest use.

Pro is enough to try this. A team that checks in every hour uses a lot more, so plan on Max once you lean on it daily. Prices change, so check the page before you sign up.

How to build your own, step by step

This is the part I wish someone had handed me. Plan on one afternoon for steps 1 through 6, then a week of small fixes.

Step 1: Pick your plan

Sign up for Claude Pro to try it, or Max if you already know you'll use it every day. Use the same login on your phone and your computer.

Step 2: Install Claude Code

You have two easy choices:

• The desktop app. Download the Claude app for Mac or Windows, sign in, and open the Code tab.
• The terminal. If you're comfortable typing commands, follow the one-line install in the Claude Code docs. Then type claude to start it.

Either way works. I do my setup on my laptop and my day-to-day from my phone.

Step 3: Make your local environment

"Local environment" sounds fancy. It just means a folder on your computer where Claude Code works. Think of it as the team's office. Every file it reads, writes or keeps lives in that folder.

Make a new folder, something like my-business. Open Claude Code in that folder. Put in anything you want the team to know, like a price sheet, a list of your properties, or your past newsletters.

Want the team to keep working when your laptop is closed? Put the folder in a private GitHub repo. Then Claude Code can open it in the cloud, and you can check on it from the Code tab in the Claude phone app. You can skip this on day one.

Step 4: Write your CLAUDE.md

"MD files" are plain text files whose names end in .md. You can open them in any text editor. Claude Code reads a file named CLAUDE.md at the start of every session. That means you explain yourself once, and every session starts out knowing you.

Type /init in Claude Code and it will draft a starter file for you. Then make it yours. Here's the kind of thing that goes in:

# About me
I'm a real estate investor. I run rentals, two apps and a blog.
I work from my phone. Keep everything short and plain.

# My businesses
- Rentals: long-term and short-term
- Apps: tools for real estate investors
- Blog: one article a week

# Rules
- Never send, post, pay, buy, delete or go live without asking me.
- Business emails stay business only.
- Never guess a fact. Find the source or leave it out.
- Finish the work. Don't hand me a to-do list.

Keep it to a page or two. You'll add to it every week.

Step 5: Write your chief of staff's playbook

This is the file that turns Claude Code into a chief of staff. Make a file called playbook.md in your folder and tell it what to do on every run:

# Chief of staff playbook
You are my chief of staff. Every run:
1. Read CLAUDE.md, mistakes.md and the board.
2. Read new email, my calendar and any notes I left.
3. Turn each new ask into a task. Give it to the right
   helper in .claude/agents/.
4. If no helper fits the job, write a new helper file.
5. Check the work. Send anything weak back for another round.
6. Put finished work on the board as a card with one button.
7. Write me a 5-line brief: done, moving, and the one
   decision I need to make.

Then add one line to your CLAUDE.md: "When I say 'run the playbook', follow playbook.md."

Step 6: Connect your apps

Your chief of staff can only see what you connect. In Claude, these are called connectors. Turn on the ones you live in, like Gmail, Google Calendar, Google Drive, Slack or QuickBooks. You'll find them in your Claude settings under Connectors. In Claude Code, type /mcp to see what's connected.

Start small. Email and calendar alone cover most of a week.

Step 7: Let your chief of staff hire the team

Here's where I didn't do the hiring. Matt did. Tell your chief of staff something like: "Look at my last two weeks of email and calendar. Decide which helpers you need, and write a job file for each one."

In Claude Code, each helper is a subagent. A subagent is one .md file in a folder called .claude/agents/ inside your project. Type /agents to see them. Here's what one looks like:

---
name: ivy
description: Inbox helper. Use for email replies and follow-ups.
---
You are Ivy. You handle the inbox.

Goals
- No email that needs a reply waits more than a day.

Every run
- Read new email. Draft replies. Never send them.
- Put each draft on the board as a card to approve.

Rules
- Read CLAUDE.md and mistakes.md first.

The top part, between the two --- lines, is the helper's name and when to use it. Everything below is its job description. Leave out a tools line and the helper gets the same tools as the main session.

You don't need 18 helpers on day one. Start with the chief of staff. Let it hire as the work shows up.

Step 8: Give it one place to put finished work

Without this, work gets scattered across chats and files. Ask your chief of staff to build you a simple private page: a short brief at the top, then a list of cards. Each card says what you asked, what the team did, how they checked it, and has one button.

Mine is one private page inside Claude. A shared Google Doc or a simple board.md file works fine to start.

Step 9: Put it on a schedule

A chief of staff that only works when you remember to open it is just a chat. Claude Code can run a saved task on a schedule, so the playbook runs without you. Mine runs every hour during my workday. Start with once in the morning and once in the evening.

Step 10: Set hard rules

Write your hard rules in CLAUDE.md and repeat them in the playbook: never send, post, pay, buy, delete or go live without your OK. Claude Code also asks before it takes actions you haven't allowed, and you can review those settings anytime with /permissions. Leave those prompts on until you trust the team.

Step 11: Make every note a rule

This is the part people skip, and it's the part that matters most. When something comes back wrong, leave a note on the card. Have your chief of staff add each note to a file called mistakes.md that every helper reads before it starts.

On day three, about four in ten of my cards needed a note from me. Wrong state on an address. An email I didn't want sent. Each note became a rule, and the same mistake stopped coming back.

The honest limits

It's not magic, and I want you to go in with clear eyes.

• It still makes mistakes. That's why the checkers and the notes matter.
• Some websites block it. When that happens, the team tells me what they couldn't check instead of guessing.
• It can't sign up for things, pay bills or press buttons that need my login. Those come to me as one link with the exact steps.
• It knows only what you connect. If an app isn't connected, the team can't see it.

And the big one: you're still the boss. The team does the work. You still make the calls.

Why this matters to me

I've written before about running my rentals from an RV and about building software without code. This is the next step. It's not about doing more work. It's about getting the work done without giving up the hours that matter most.

On a normal evening now, I'm in the yard with a soccer ball at 6 PM. Meanwhile, a page gets tested, an email gets written, and an insurance quote gets compared. When I pick up my phone after dinner, it's all waiting for one tap.

If you've been stuck between the life you want to build and the life you want to live, this is how I got unstuck. Build your chief of staff. Give it one job. See how much time comes back.

Jennifer