← Back to posts

I hired an AI chief of staff for a $0 salary

I counted 55 chats in my Codex history. There's a teleprompter app in there, a plan for testing paid ads, a website that went down, even one about why my Mac ran out of disk space. Looking at that list, I had a pretty good idea why opening Codex was starting to feel overwhelming.

You can do damn near anything with AI. That's fun for me right up until I sit down with half an hour to work and spend it figuring out where I left off. I open the homepage chat to look at a mockup, see a question about ads on the way there, and remember I still owe you guys a newsletter. I could start any of those in seconds. Which one am I actually going to finish today?

So this week I hired a chief of staff. It's another Codex chat, on the subscription I already pay for. Salary is $0. Its job is to remember what I've committed to, follow up with the agents doing that work, and bring me the handful of things that need my attention.

Which chats get a follow-up?

I started with my product feed tool, paid ads, and this newsletter. Those are three places where I regularly owe somebody an answer or expect something back. The rest of my history can sit there. A chat where I explored an idea two weeks ago doesn't suddenly need a deadline because an agent found it.

Here's how I'd decide what to give yours. Say you have a chat with ten ideas for ads and another where you asked for two cuts of a video you want to test Friday. I'd give the chief the second one. Tell it what you're waiting for and when you want to see it. Leave the ideas alone until you've picked one. You can add another chat whenever there's something in it you want followed through. That one choice saves you from getting a nicely formatted list of 55 things to feel behind on.

Today there was a small decision about the homepage for the feed tool. I liked headline number two. The chief could see I'd picked it, found the version of the mockup where the agent had actually used it, and took the headline off my list. The layout still needed work, so it kept that part open. That's about as exciting as managing a homepage edit sounds, which is exactly why I wanted somebody else doing it.

What I want to know in the morning

At 8:30 Central, my chief checks the chats I've put on its list. I gave it 180 words and room for up to three things. If an agent needs a decision, tell me what it recommends and give me the link I'd need to make the call. If something I asked for is ready, show me where I can use it. If nothing changed, let me have my coffee.

I can answer in that same chat and the chief takes my answer back to the agent doing the work. During the day it checks in hourly while my Mac is awake and Codex is running. It keeps a little list of who owes what, when to check again, and what would let us call it done. If an agent is already working, it leaves it alone. Two unanswered follow-ups and it brings the blocker to me.

One distinction I care about as a marketer is what counts as permission. I might ask the ads agent to check spend this afternoon. The chief can chase down the numbers and bring me a recommendation. Raising the budget is still my call. Same with the newsletter. Asking if the draft is ready doesn't give anybody permission to hit send.

The prompt I'd start with

Open a new chat in a setup that can read and message your other agents. Pick a few chats where you're waiting on something and give it the links. Here's a version you can copy and fill in.

Act as my chief of staff for these chats: [chat links or names]. First check which ones you can read and message. If you can't access one, tell me what connection or pasted update you need.

Start by asking what I want to finish this week. Track the work I've committed to in those chats. Leave brainstorms parked unless I pick them up. If I add a chat later, ask what outcome I want you to follow. Read the latest work in each chat. Keep a running list of open commitments, who owns each one, the next step, when to check back, and what would prove it's done. Show me the first brief now. Include up to three things worth my attention in under 180 words. For a decision, give me your recommendation. For finished work, give me the link or result.

Follow up in the named chats when something is due. Ask for the result and what's still open. If I reply here, carry my answer back to the right chat. Give an agent time to work. After two unanswered follow-ups, bring the blocker to me.

Keep my existing approval rules. Ask me before spending money, publishing, sending anything outside these chats, or expanding the work. Once I've reviewed the first brief, help me set a daily time in my timezone. Tell me what needs to be running for that schedule to work.

I'd run it once before setting up the schedule. Look at the first item it brings back. Could you answer it right there, or would you still have to go read 40 messages to figure out what the agent wants? If you're opening the old chat anyway, show the chief what you had to go find. That's the detail you want in tomorrow's brief.