• D.C.
  • BXL
  • Lagos
  • Riyadh
  • Beijing
  • SG

Intelligence for the New World Economy

  • D.C.
  • BXL
  • Lagos
Semafor Logo
  • Riyadh
  • Beijing
  • SG


Agentic brains and ‘digital gardeners’: How one CEO runs his AI office

Andrew Edgecliffe-Johnson
Andrew Edgecliffe-Johnson
CEO Editor, Semafor
Nov 7, 2025, 5:00am EST
CEO SignalBusinessNorth America
Illustration of Jim O’Leary
Joey Pfiefer/Semafor
PostEmailWhatsapp

This article first appeared in The CEO Signal. Request an invitation.

Title icon

The Signal Insight

As North America CEO of one of the world’s largest communications agencies, Jim O’Leary was used to clients asking him how AI could improve their PR and marketing.

Earlier this year, though, he noticed the conversation among CEOs shifting to what the technology could do for their own productivity.

His response was to ask his team to build him “a digital version of my brain” — a personal AI workspace that turns ideas into drafts of memos, crunches down long presentations, and keeps him abreast of what the competition is up to. The system saves him an hour or two a day to do work that is “more creative, more strategic, and more interesting,” he says. He expects most CEOs at large companies will soon be using something similar.

There are caveats, O’Leary says, notably that sensitive data must be kept in secure environments. But as AI threatens many white-collar jobs, O’Leary says he doesn’t fear being replaced if his clients adopt such tools themselves.

“What you get from an agent is speed and scale. And what you get from a human is judgment, creativity, and intuition,” he says — and any client will ask: “Am I going to be content with something that is potentially better speed and scale, or would I prefer something that’s speed, scale, judgment, intuition, and creativity?”

AD

This interview has been edited for length and clarity.

Andrew Edgecliffe-Johnson: You talk about turning your office into a test bed for AI. What does that mean?

Jim O’Leary: [I told] my team, “If I’m going to be talking to other CEOs about this, I would like to make sure that I have the most sophisticated personal AI system that can be built.” [They replied:] “First, we need to interview you, to determine from the moment you get up to the moment you go to bed, what you are doing, and break everything that you do down to the atomic unit level.” [They did that, and said:] “These tasks here can be automated with AI, these tasks can be supercharged with custom agents, and these tasks here can be improved with off-the-shelf tech that already exists.” My custom system [feeds into] a centralized place to organize everything. It’s like a digital version of my brain or my own miniature large language model.

So I could have the LLM go through my emails and pull together whatever I need. I have it doing automated note-taking and summarizations for meetings and discussions. I also have a custom suite of agents, [including] a writing agent and an intelligence-gathering agent, so I get real-time competitive intel being fed to me, and agents that are more strategic. So if I want to bounce ideas off a version of myself that is built to challenge me, then I have that.

What are the off-the-shelf solutions missing in terms of what’s most valuable for someone in your role?

I call it the 90:10 paradox, where a lot of CEOs are quite accustomed to getting things ready for prime time, and AI doesn’t exactly work that way. You can take a bunch of off-the-shelf tech, put it in a secure environment, layer it, have it talk to each other, but it still is only going to get you to the 90-yard line, if you’re talking American football. And the real magic happens in the last 10 yards. That’s where you score the touchdown. I think a lot of CEOs might have something that gets them to the 90-yard line, and they’re unhappy with it because it doesn’t live up to the hype.

AD

Which features are you finding most valuable?

I have found that if I have a perspective or an insight that occurs to me in the moment, I can feed that into my writing agent and have it turn into something more than just a random thought. It could turn into the first draft of an email for me to send to the company. It could turn into the first draft of something for me to put on LinkedIn. It could turn into the first draft of something for me to share with my CEO friends or clients.

A much more interesting example is competitive intelligence. [You can ask an agent] to identify your competitors. What are they doing? What does it mean for you? Then have your agent [offer] its suggestions — is there anything you should do? Maybe you want it fed to you in a way that, on your morning jog, you can listen to the entire thing in the equivalent of a podcast format, in your own voice.

Have you done that?

Yes, this morning. Sometimes I will also do that if I have a very large set of briefing materials that I need to go through, and I’m traveling. If I’m going in and out of the airport, I just put my headphones on, I put the raw materials in, and the whole time [from arriving to] getting on the airplane, I’m getting the audio briefing.

AD

If you’re a CEO, you get an enormous amount of information. People are presenting plans and strategies and ideas and pitching you. And traditionally, we would get a pre-read, and we would read through it all and maybe make some notes and have some questions. And then we’d jump on the call, and someone would present to us. Now I have my agent typically take the first stab at ingesting the full packet of information, and then make recommendations to me on certain things that I should explore and unpack. It skips the first and second steps so that I have more mental energy for the important step, which is the meeting itself.

So you brainstormed this with your team, the office of the CEO. Has it replaced them all?

My supporting office is, actually, one person more than it was before. It just does a lot more. The role we’ve added is our agentic leader, who has been building agents for the past 24 to 36 months. She’s AI-native, and her entire job is to be the digital gardener for what essentially is our AI garden. All the agents need to be watered. They need to be fertilized. They need to be refreshed.

All the project knowledge that you have loaded in needs to be updated when it becomes outdated, or when you create new best-of examples of everything you’re doing. You want to load those in and push out the previous best-of examples. You don’t train your system based on the mediocre speech you gave a month ago. You train it exclusively on the best speech that you gave or the best strategy presentation that you’ve created.

How have you cascaded this through the organization?

We are doing a month-long training for every single member of my [executive leadership team]. They will each get their own individual system within our broader system, then they will continue the cascade. The goal with a transformation like this is to cascade it level by level, so that you can make sure that it’s not like one of those age-old corporate initiatives that is a bunch of slides and talk but doesn’t actually have any practical application.

Title icon

Notable

  • AI agents are getting a lobby group in DC, Semafor’s Reed Albergotti reported. Members of the new Agentic Futures Initiative believe lawmakers and officials need to better understand the technology to ensure new products remain interoperable across platforms, private — and secure. Agents need to access different software, and those interconnections create vulnerabilities.
  • Some junior executives are creating a different type of CEO agent. Alex Alonso, the chief data and analytics officer at the Society for Human Resource Management, made an AI chatbot of Johnny Taylor Jr., his boss, to offer feedback on ideas before he presents them to the real Taylor. The results have been a “work in progress,” Alonso told CNN.
AD
AD