What an AI chief of staff actually does all day

"AI chief of staff" sounds like a job title someone made up for a pitch deck. I thought so too, until I built one and noticed how much of my day it quietly took over.

Mine is called Edison. He runs on my own PC, talks to me out loud, and works for one founder running two businesses. This post is the honest job description: what he actually does from morning to night, what he is not allowed to do, and how you could set up something similar.

A chatbot waits. A chief of staff doesn't.

The difference is simple. A chatbot sits in a tab until you think of a question. A chief of staff already knows what's going on, tells you what matters, hands work to the right people, and only bothers you when a decision is yours.

So the test for an AI chief of staff isn't "can it answer questions". It's:

• Does it know what happened while I wasn't looking?
• Can it hand work to someone else and follow up?
• Does it protect me from mistakes, including its own?
• Does it remember what we decided last week?

Here's how Edison handles each of those.

The job list

1. Keep watch, every hour

Every hour from 7 AM to 10 PM, Edison does a quiet read of my world: my inbox, my calendar, my notes. He's only allowed to look, not touch. What he finds goes onto small tiles on his board, so when I glance over I see what changed without opening anything.

2. The morning brief

When I sit down, he gives me a short spoken briefing built only from those real numbers: meetings today, approvals waiting, what came in overnight. If a number isn't there, he doesn't make one up. Twelve hours of his own journal also go into a catch-up, so he knows what I was doing yesterday.

3. Hand out the work

Each of my businesses has its own AI team: a "CEO" with a to-do list and a few specialists. When I say "have the GlowBop team check the launch list", Edison hands the job to that team. It opens as a chat in my Claude app, so I can watch or step in. When they finish, a soft chime plays and he reads me the report.

This is the part that changed my day the most. I stopped being the person who copies tasks between ten AI chats. I just say what I want, to one voice.

4. Guard the money and the megaphone

Anything that spends money, posts in public, sends a message or deletes something waits on an approval card: what it is, what it costs, and the proof it's a good idea. Nothing happens until I tap Approve or say yes to his read-back. There's also one key that stops everything at once.

The teams are allowed to ask for spending. Only I can say yes. That one rule is what makes it safe to let an AI do real work.

5. Remember everything

Every conversation, answer and job result goes into a journal on my computer: a note per day, then weekly, monthly, quarterly and yearly summaries. So "what did I ask you yesterday?" has a real answer, and so does "what did we decide about the launch date?".

6. Run the house (carefully)

He reads my smart home all day, but he can switch exactly one thing: the TV. "Turn on the TV and open YouTube" works. Everything else in the house is look-only until I decide otherwise, one device at a time.

7. Look after himself

A chief of staff who breaks quietly is worse than none. Edison checks his own voice, memory, disk and connections, says a new problem once, in one sentence, with one fix, and does the safe fixes himself: restarting a slow voice engine, closing leftover test browsers, setting aside giant log files.

What he is not allowed to do

This list matters as much as the one above:

• He doesn't send email, post or spend on his own. Ever. Approval first.
• He reads my mail, calendar and files; he doesn't change them.
• The AI teams can't approve their own work: the key that approves things lives outside the folders they work in, where their tools can't reach it.
• He doesn't chatter. Quiet sounds tell me he heard me, he's thinking, or the answer is ready.

Why voice changes everything

You can build a lot of this with dashboards and chat windows. What makes it feel like a chief of staff is that I can talk to him from across the room while my hands are busy, and he talks back. And because I can talk over him mid-sentence, it feels like a conversation, not a voicemail.

How to start your own (with or without Edison)

If you want an AI chief of staff, these five steps are the ones that mattered for me:

1. Write down what you check every day. Inbox, calendar, sales, a few dashboards. That's the hourly watch list.
2. Start read-only. Let it look at everything and change nothing. Add powers one at a time.
3. Put one gate in front of money and public posts. Cost, proof, and a human yes.
4. Give it a memory. A plain daily journal is enough to start.
5. Give it one voice. One assistant you talk to, which hands work to others, beats ten chats you juggle.

Meet mine

You can wake Edison up on the home page and talk to him. That demo runs offline with made-up data, but it's his real voice and sounds. If you'd like your own, sign up to get Edison: you get him first and help decide what he learns next.