I run a show with a team that doesn't exist. Here's the org chart.

At the end of 2019 I had two production quotes for a show I was trying to start. The smaller shop wanted $1,400 an episode. The other, with three cameras, a sound engineer and lights, came to $2,955 to $3,000.

It was called The Atlanta Table. Once a month I'd sit down for lunch or dinner with people running consumer businesses in Atlanta, and we'd record it as a web series or a podcast. I'd already got the restaurant and the food covered for free, so every dollar on those quotes was production: people to shoot it, someone on sound, lights, setup and breakdown, fifteen hours of editing. For a show that was meant to come out every month, that was a budget decision before it was ever a creative one, and it never got made.

The first episode of The Operator Economy went out at 11 this morning. It's a different show, but it needs the same list of jobs. Somebody finds the story, somebody researches it, somebody writes it, somebody voices it, somebody appears on screen, somebody builds the parts nobody filmed, somebody edits. I haven't hired any of those people. I've given each job to an AI tool, and kept a short list of calls I don't hand over.

Put the two side by side. The Atlanta Table was one lunch in one room, and it was quoted at close to $3,000 an episode. The Operator Economy needs research, a script, narration, a presenter, animation and generated footage. I had each of those jobs priced at published U.S. rates, mostly government wage figures and the actors' union rate sheet, with cautious guesses at the hours. One episode comes to roughly $8,800 at the low end and $15,600 in the middle, and more than half of that is animation. It's a rough estimate, not a quote, and real bids would probably come in higher.

In 2019 the cost arrived as a quote for every episode, before I'd made a single one. What changed is that I didn't need that money up front before I could make the first one.

One of those jobs is the presenter. The person presenting the show isn't me; it's an AI presenter built from my likeness.

Here's the whole setup, in the order an episode moves through it.

Finding the idea

A scheduled job runs four times a week. It searches the web and public forums for problems somebody would pay to have solved, turns the strong ones into possible businesses, and keeps a shortlist of up to five. Once a week a second job takes one lead off that list, writes it up properly with sources, and argues against it. It isn't allowed to approve anything. That's how it works now; the first episode came from an earlier version of the same idea, where AI drafted the research and I did the same reading. My part is the argument. I read what it found and push on two things: would this business actually work, and is there a story in it anyone would sit through.

For this episode, the research threw out the idea's own headline claim, the kind of number that makes a pitch, because it was published by firms that sell the service. The most persuasive line in the whole idea didn't survive the first honest read.

Writing it

Claude Code runs as an editorial team. One role works out the business model, one keeps every claim inside what the research supports, one builds the story, one writes the script, and a review pass exists only to attack the draft. Each objection gets a written answer: fixed, changed, rejected, or kept on purpose, with the reason. Then I read the final script once, straight through, and answer one question. Does this sound like me talking? Nothing locks until that's a yes.

It still got something wrong. Three weeks after the script locked, I cut a line about my own background that wasn't true. Now that part of the episode opens with "I have no transaction experience," which is less flattering and correct.

The voice

Google's text-to-speech performs the script, and ElevenLabs turns that performance into a clone of my own voice. So it sounds like me, but I didn't record a word of it. My job here is mostly saying no. While setting up the voice, I rejected the first candidates for sounding flat, with no life in them.

The picture

Scenes are built in code with a tool called HyperFrames. Footage the script needs but nobody filmed is generated with Kling. The presenter is made in Higgsfield. I review every round, and there were a lot of rounds.

Finishing and shipping

The cut and the sound come together in the free version of DaVinci Resolve. A script uploads the episode and its Shorts to YouTube and schedules them. The title and thumbnail calls are mine.

How it's governed

Codex and Claude Code do nearly all of the building, so I spend my time deciding what each one is allowed to do, not which one to use.

Most stages run from a written standard I've approved; the rest are being written from what this episode taught. Between the early stages there's a gate: a list of what has to be true before the work moves on. The main rule is written into the gates themselves. No automated check can approve anything. Checks can find problems and prove the pieces agree with each other; only I approve what an episode says.

When a script locks, its files are fingerprinted, so any change after that has to be made in the open and logged. That's how the line I cut three weeks later shows up in the record.

My decisions go into a log in my own words, tied to the exact file I was looking at. That's how I know which version of a scene I approved, and why.

The scheduled jobs have their limits written into them. The scout can't promote an idea. The research job can't approve one, assign it an episode, or spend money. Anything that costs money waits for my yes, and every paid call gets logged.

And the first episode left runbooks behind, four of them, for finishing, packaging, Shorts and LinkedIn, so the next one starts from what this one learned.

What's left

If you took the tools away, what's left is a list of decisions: which idea goes forward, which words get said, which take is good enough, what gets cut, when it ships. It's the part I'd have been worst at delegating even if I'd had the budget for a real team, because it's where what I know about businesses and about my own voice actually does something.

It's the first episode. It went out an hour and a half ago, and I have no idea yet whether anyone wants it. The earlier version of this show had no presenter and no generated footage; this one does, and I'm starting the count again from one. So don't read this as advice on growing a channel. It's how one person now makes something that used to take a room full of people, and what's still mine to do.

Here's the first episode: https://youtu.be/7juZ1KXdd04