25 Things on a Whiteboard: What a Fleet of AI Agents Taught Me About the Founder Bottleneck

On March 29 I did an exercise I’d been putting off. I put 25 things on a whiteboard: everything stopping me from stepping away from my own company.

Then I traced each one back to its source.

Every single one came back to me.

That same day I posted something that sounds like the opposite. I no longer thought of having computers. I had compute and storage, and I could treat all my devices and computers as one personal fleet, managed by an agent I call Sys3P0. I already had an AI chief of staff. This felt like nirvana.

Both things were true on the same day. That’s the most honest thing I can tell you about running a company with AI agents: you can automate a lot of the work and still be the bottleneck.

The founder bottleneck isn’t a flex

Founders love being needed. We call it “being close to the details.” What it usually means is that decisions live in one person’s head, and that person is out of bandwidth.

Andy Grove put it plainly in High Output Management: a manager’s output is the output of the organization they run, plus the organizations they influence. It isn’t measured by how many things the manager personally touches. By Grove’s measure, a CEO who appears in 25 out of 25 blockers isn’t high output. He’s a very busy single point of failure.

I know this. The whiteboard still came back with my name on every line.

What the fleet actually is

This isn’t a sci-fi butler, so let me be concrete.

At the center there’s a chief-of-staff agent, Yobi-1. It sits on top of a private notes vault: my brain dumps, plans, book notes, and the context I’d otherwise carry around in my head. It turns that raw capture into tasks and plans, and it keeps my private thinking separate from what the team sees. Sys3P0 looks after the machines.

In January I wrote that I’d been running hundreds of AI agents in my private stack. Not demos. Real systems doing real work.

The turning point wasn’t a smarter model. Last December, while people were away on break, I finally got good memory set up on my agents, along with real-time communication between them. It was another world. Even Claude Code acted differently.

Four rules I’d give anyone building one

1. Private stays private. My chief of staff is inward-facing. It holds my unfinished thinking, and it doesn’t broadcast. When the line between “my thinking” and “what the team sees” blurs, trust breaks fast.

2. Memory is the product. A chat window forgets everything by Friday. An agent that remembers your decisions, and why you made them, compounds. The agent is the easy part. What it knows is the moat.

3. Agents should protect attention, not consume it. In early 2025 I wrote that my AI chief of staff uses ideas from Nir Eyal’s Indistractable to keep me focused. That’s still the test. If an agent creates more pings than it removes, I’ve built a slot machine, not a staff.

4. Write the decision down, or no one can take it. An agent can only act on what you’ve made explicit. A decision that lives only in your head can’t be handed to anyone, person or agent.

What the fleet didn’t fix

The whiteboard said it plainly: every one of the 25 came back to me. No agent can inherit a decision it has never seen.

The fleet gave me leverage. It didn’t give me permission to stay the bottleneck.

The question I’d leave you with

If you made your own whiteboard tonight, how many of the 25 would come back to you? And how many of those have you ever written down?

Ahmed Reza is CEO of YiB Inc., the home of Yobi. He writes about building with AI, books, and the people both are supposed to serve. More about Ahmed.