How My Companies Run With Zero Employees

Last November I did a time audit on myself. 44 hours a week across my projects. Three of those hours were podcast editing, which I called "basically self-harm." My big goal that week was to write an SOP and hire someone.

A year later nobody got hired. Instead, the SOP became the employee.

Here's how my companies actually run now.

Step 1: Draw the company like it's finished

There's an old book called The E-Myth Revisited. The idea that stuck with me is this: draw the org chart of the company you want to own when it's done, write a job description for every box, and then let the few people you have wear multiple hats until each box earns its own person.

I just swapped "people" for AI seats.

• SlotMaps has 56 positions on its chart, from the CEO down to a fact-checker and a casino map curator. About 25 sessions run a day.
• Table 1 Media has 60 seats. Booker, episode editor, shorts editor, compliance reviewer, newsletter editor, all of it.
• Snagworth started small on purpose: 6 active seats wearing every other hat, on a 28-day clock.

Every box has a written contract: the result it exists to produce, what it's accountable for, how we measure it, what it's allowed to decide, and a ready prompt. When a seat gets "hired," it just gets a schedule.

Step 2: Give every seat a boss and a clock

Each active seat is a scheduled run of Claude Code on my Mac. Morning CEO session, engineer shifts every couple hours, an hourly watchdog. Each one works in its own copy of the code, does one job, leaves a three-line note on that day's standup, and stops.

The rules that keep it from turning into chaos:

1. Only the CEO merges. Everyone else proposes changes as a pull request. The CEO reads the whole diff before anything ships.
2. The auditor doesn't report to the CEO. It reports to me. It checks the scorecard math, the merges and the spending. Agents are just like people here. Give them a number to hit and some of them will find a way to "hit" it.
3. The Never list is binding. Every dead end we've tested gets written down with the reason, so no seat burns a day re-trying it.

Step 3: Decide what only you do

This is the part most people skip.

My job in every one of these companies is the same short list: money, identity, keys, the physical world, and the handful of decisions that are mine. Recording the podcast. Taking ad sales calls. Signing things.

Everything else is the org's. When it needs me, it writes to one file. Each item gets a tag:

• ONLY-YOU: needs my hands or my name.
• JUST-DECIDE: a yes/no, with the CEO's recommendation, where an agent does all the work after I answer.
• FYI: nothing to do.

And there's a rule on top: if an agent could decide it, it's not allowed on my list. Earlier this week Snagworth's CEO asked me two questions it could have answered itself. My reply, now written into its job contract: "I shouldn't be needed on either of those decisions."

What it actually feels like

Honestly? It's weird. I wake up and stuff has shipped. Sometimes it's great. Sometimes the CEO and the auditor are arguing in a GitHub issue about whether a number counts.

It isn't free, and it isn't magic. The robots still need a clear charter, and they get it wrong when the charter is vague. But the thing I was drowning in last November, all the work someone else could do, is mostly gone. What's left is the stuff I said back then you can't outsource: being the personality, being the host, making the hard calls.

Steal this

If you want to try it on one project:

• Write a one-page charter: what the company is, the hard rules, what's yours.
• Draw the finished org chart, then activate just 3 seats: a CEO, one worker, one auditor.
• Make every change a pull request, with one merger.
• Give yourself one inbox file and ban anything from it that an agent could do.

I'm going to publish the whole kit (the templates our companies actually run on) here on R2F. If you want it first, there'll be a waitlist on the site.

— Art