Think Like a CEO: What 10 Years as a Chief of Staff Taught Me About Managing AI Agents

Almost everything I learned about managing agents I learned from 10 years as a chief of staff working for CEOs and executives.

You see, CEOs are a different breed of people to operators.

Operators live in the detail. We’re the ones in the tools, building the systems, writing the briefs, checking the work. We know how everything gets done because most of the time we’re the ones doing it.

CEOs spend their day deciding what matters and putting the right people on it. They hold the big picture, the strategy and the direction of the business. Their attention goes to the few things only they can do, and everything else belongs to somebody else.

When I decided to go all in on building my one person business with AI as my only employee, I realised quickly I needed to act less like the operator and more like the CEO. And I’ve been able to transform my whole business in the last year just through this mindset.

Here are five things I learned from CEOs that have helped me turn my business into a one person + AI powerhouse:

1. They talk, don’t type

Every CEO I worked with operated primarily by talking.

They didn’t open an app, create a document and carefully type up a brief every time they wanted something done.

If I was lucky it was a phone call or zoom. Most of the time it was a rambled voice note on Slack at 9pm at night.

They talked. Things happened.

And now that’s how I work too. Despite being able to type 140+ words a minute, mostly I communicate now through voice.

• I use Wispr Flow in AI chats to send messages to agents.
• I dictate briefs, documents and any other content that needs to be down on paper. My agents clean it up.
• I’ve been experimenting going full voice mode with ChatGPT Voice and other voice agents - where I just talk all day.
• I even capture everything with voice now. My whole second brain is voice notes that my agent goes and organises.

Voice just removes a huge amount of friction from the work. I don’t have to be at my laptop. If a thought comes into my head I can just get it out, get it delegated or filed away and I know it’s taken care of.

2. They aren’t on the tools

I spent years trying to get the CEOs I worked for to use tools to organise themselves. Notion, Trello, Monday, various CRM tools, etc.

Almost none of them ever logged in to the systems I built.

If they needed something… they asked (remember point 1).

They are just not built to click around in systems and interfaces. And they have long since been off the actual operational tools. And it’s not that they are lazy. Their time, effort & attention is on other things… and rightly so.

And that is sometimes hard to understand if you’re an operator. I have always been the one on the tools. I’ll design the landing page, setup the lead magnet, run the Facebook Ads. I’m a generalist. I’m pretty good at most things.

But that also means I can go down rabbit holes when I shouldn’t.

Over the past year working with AI, I have gradually moved off the tools in my own business too. My agents have access to every tool in my business and can operate them just like I could in the past.

All the stuff I used to do myself, my agent does:

• Uploads & sends emails in Kit (my email marketing software)
• Updates my website + landing pages
• Does almost everything in Circle my membership community including creating posts, upgrading/downgrading members, creating courses, etc.
• My agents even do all my video editing for me now.

All the fiddly stuff that used to take up all my time… I’ve been able to give that to AI to take care of.

3. They care about the outcome not the steps

I remember as an operator when I first started delegating work. I would write up insane briefs, give people every step they needed to complete the work. By the time I was done, I probably could have just done the work myself.

I thought this was how you got the best work back when delegating.

Compare that to how I watched CEOs delegate. They tell someone what they want, with very little context or information and trust that the people they hired are smart enough to make it happen.

And to my surprise. Their way worked WAY better than mine. People came back with results that were surprisingly good because they weren’t constrained by micro managed steps.

And it turns out the AI models today work in a very similar way. They are designed to be smart enough that you can give them a goal and they will work out how to get it done.

This means you need to get good at articulating outcomes, goals and what good looks like rather than writing step by step instructions.

Every skill I build now has fewer steps and more goals, context & why. The big picture stuff. I include three sections in skills that help with this:

The Goal: What is the end result we are trying to get to with this skill. The Why: Why this matters to ME. I find this gives my agents better judgement. If they understand why it matters, they do much better work. What Success Looks Like: Ultimately what the end result you are looking for is.

And this doesn’t just work for skills you run. It can work in any conversation. Quite often when I feel myself getting into a spiral loop with AI, going back and forth and not quite getting anywhere, I will just stop and rearticulate the goal again. And just tell it to go work it out. 99% of the time that works to break that loop.

4. They take the time to train

I had one boss that would make me write all my own SOPs as he was training me. He would teach me something, I would do it a few times. Then I would have to write the SOP.

It was gruelling, but also turned out to be a fantastic way to learn. Writing an SOP makes you take notice of every step and embeds it deeply in your memory.

I’ve used the same method when training human employees and now my own AI Agents. I call it the Codify Method:

1. Do the work together
2. Ask your agent to create a skill so we can do this again
3. Use the skill next time you need to repeat the work
4. Iterate on anything that didn’t work.
5. Repeat until it runs seamlessly.

Over the last year I’ve built over 50 skills for all sorts of workflows in the business. And now my agents know how to do most of the business activity without skipping a beat.

By the way, I created the Codify Your Work Toolkit that helps you map out your business workflows and teach your agents how to create skills that are repeatable. You can grab it here.

5. They don’t get mad. They just fix things

I remember when I was a VERY young receptionist just starting out I made a really bad mistake. I hadn’t confirmed a meeting for my boss in which he had to drive an hour to a client’s house and they were not there.

I was SO scared my boss was going to come back and fire me. But when he came back to the office he brought me into his office and calmly said, this was a bit of a stuff up Ev, how are we going to fix this so it doesn’t happen again?

That was it. He didn’t yell. He wasn’t irate. He just wanted to know it wouldn’t happen again.

Now I don’t know about you, but I have had my fair share of yelling at AI for doing something wrong, or stuffing something up. But I’ve learned none of that is worth my time. My agent doesn’t care if I yell and scream, it doesn’t have feelings, it won’t ‘do better next time’ unless we fix the problem.

If you don’t document mistakes and corrections and preferences your agent will have NO clue for next time. So my one rule that is in my claude.md file is this:

FIX IT AT THE SOURCE.

If we are working in a skill, it looks for how we can fix it so it doesn’t happen again. If it’s persistent we’ll add it to memory, or a project file. Everything gets documented somewhere so mistakes don’t happen again.

What it actually looks like to work like this

All of this together = a totally different way of working as a solo operator until you get days like this:

I was on the way to the hairdresser which is about an hour away from my house. Someone emailed saying they couldn’t access a course they had just bought. I saw the notification pop up. So I just opened a chat with my agent and said ‘can you check the email, make sure they have the access they need and draft an email back.’

Two minutes later it was done while I was still driving.

That’s what it’s like to operate as a CEO with AI in your business.

— Ev

---

PS. Whenever you’re ready to operate like a CEO with agents running the work, here are three ways I can help:

1. Start building your very first agent in your business with the AI Chief of Staff Course. This is how you go from operator to CEO in the shortest amount of time.
2. Upgrade your agent by building out a second brain with the AI Second Brain Toolkit so your agent has the context it needs to operate with you in the business
3. Or go all in with my membership, The College of Knowledge. Where we build your full personal agent operating system. Get ongoing support, office hours, toolkits and a community of other solo operators to do it with.