The CEO-Mode of Working in the AI Era

I recently realized something: I haven’t written code by hand in a while.

Not because I can’t. Because I don’t need to. Since switching my daily workflow to Claude Code, my role has fundamentally changed — from an Individual Contributor to something that looks a lot like a CEO.

What a Typical Day Looks Like

Every day when I open my laptop, here’s roughly what I do:

1. Assign tasks. Kick off research with Wide Research, or describe a requirement to Claude Code.
2. Review proposals. Read through the agent’s initial findings or execution plans (Plan mode). Decide if the direction is right.
3. Approve or reject. Green-light it and let it run, or send it back with feedback.

Then repeat. On any given day, I might have five or six Claude Code sessions open simultaneously, each working on a different project. I rotate between them — reviewing, approving, rejecting, reviewing again.

It’s not fundamentally different from managing a small team.

Productivity Is Genuinely Amplified

Before, pushing forward one or two projects a day was a good day. Now I can run five or six in parallel, because execution is almost entirely offloaded to AI. My job is directional judgment and quality control — precisely the parts AI still can’t reliably do on its own.

Product research for e-commerce, technical implementation for content projects, writing and publishing blog posts — things that used to demand dedicated blocks of time can now run concurrently. It feels like suddenly having several competent reports who occasionally need course corrections but execute fast.

But the Fatigue Is Different

The old fatigue came from writing code, debugging, doing repetitive work. The new fatigue comes from decision fatigue.

Every result an agent returns requires a judgment call: Is this heading in the right direction? What’s missing? Should I adjust? Making dozens of these calls a day, by evening my brain simply stops working.

It reminds me of research on CEO decision fatigue — they’re not physically tired, they’re depleting their judgment capacity. Now I get it firsthand.

It looks like not writing code should be easier. In practice, the sheer volume of things I can now take on means the daily information load and decision count are higher than ever.

The New Bottlenecks

I posted this on social media recently:

The only things that can make me rest now — the only things limiting my productivity — are sleep and AI company reliability.

This isn’t hyperbole. When your workflow becomes a loop of “assign → review → approve,” you can theoretically keep producing as long as you’re awake and the AI service is up. Without the physical barrier of “too tired to code,” the ceiling on productivity becomes two things:

• Your energy. Specifically, sleep. Enough sleep means you can keep making decisions. Too little and your judgment degrades.
• AI service availability. Claude occasionally hiccups. APIs time out. Services go down. These interruptions become your forced rest periods.

In a way, AI company downtime has become my mandatory break mechanism.

A Shift Worth Documenting

I think this shift in working mode is worth writing down, because it likely represents a broader trend.

When AI coding assistants are good enough, many solo developers and small-team founders will naturally drift from executor to manager. You don’t need to hire anyone, but you do need to learn how to “manage AI” — decompose tasks, set standards, review output, control quality.

These are management skills, not programming skills.

Looking back at the past few months, what I’ve improved most isn’t any technical ability. It’s: how to describe requirements more precisely, how to evaluate proposal quality faster, how to effectively manage multiple projects in parallel. These are all core competencies of a manager.

I’m not sure if this is a good thing or a bad thing. But it’s definitely happening.