[Prologue] How I Built a $20/Month AI Content Creation Workflow

 Building an AI one-person content studio with ChatGPT, Codex, and human judgment

Turning one idea into a finished result requires more roles than it first appears. Someone has to define the problem, break the work into tasks, create the files, review the result, and decide what should happen next.

This series documents how I ran that process as an AI-agent collaboration using ChatGPT and Codex. I set the direction and quality bar. ChatGPT structured the requirements and content. Codex handled execution at the level of code, files, documents, and production packages.

The first proof case was a personal asset-management automation system. I then applied the same operating model to YouTube Shorts, webtoons, motion comics, blogs, and a future book. The result is becoming a multi-channel publishing system for an AI one-person content studio.

Why I started

The project did not begin with a grand startup idea. It began with a spreadsheet I had to update repeatedly after the Korean and U.S. stock markets closed.

I checked numbers by account, recalculated my overall position, and compared it with the previous day. The repetitive input took time and introduced opportunities for error. More importantly, focusing on moving numbers could distract me from the portfolio changes and risks I actually wanted to review.

My first goal was simple: after the market closed, update what could be updated, review the overall state, and receive a concise summary.

How the work was divided

• Human / CEO: originate ideas, define the concept, set priorities, verify facts, judge quality, and approve publication
• ChatGPT: structure requirements, design the content, review disclosure risks, and organize the next actions
• Codex: execute code and file work, assemble deliverables, package releases, and record changes

This was not a case of removing human judgment. I first defined the problem and the boundaries. ChatGPT converted that into an executable structure. Codex created and revised the actual artifacts. I then used the result, reviewed it, and decided what to change.

My current working identity

I describe myself as an AI-powered solo creator. That identity is based on completed work rather than a job title.

My current work includes YouTube, original webtoon stories, motion comics, blog publishing, and book preparation. My areas of interest include AI, automation, personal asset management, cooking, travel, and storytelling.

The first proof: a personal asset-management system

The first system was not an investment-advisory service and did not execute trades. It was a personal monitoring tool built around criteria I defined.

• a web dashboard for the overall asset flow;
• a Telegram summary after market close;
• checks for cash allocation and concentration;
• manual input for cash-like assets that were difficult to connect through APIs;
• a feedback loop that turned real-use problems into the next revision.

Organize before automating

I did not begin by writing code for everything. I first separated the numbers I checked every day, the data that could be updated automatically, the information that had to remain manual, and the rules I wanted the system to apply.

Once those decisions were clear, ChatGPT structured the requirements and Codex implemented them. Keeping difficult or low-frequency data manual was sometimes more reliable than forcing a fragile connection.

From one system to multi-channel publishing

The same collaboration pattern can be used for content. Experiences, records, photos, stories, and project outputs can be structured once and then adapted for different channels.

The target channels include long-form YouTube videos, Shorts, webtoons, motion comics, blogs, and books or ebooks. After publication, I record reach, retention, comments, and search traffic. Those signals become inputs for the next title, format, and production decision.

What “$20 per month” means

The phrase refers to the base ChatGPT Plus subscription price. Taxes may apply depending on the country; my Korean payment was about $22 including VAT.

It does not include hardware, human labor, API charges, paid text-to-speech, paid video-generation services, or other premium software. The experiment asks how far a creator can go with the base subscription at the center of the workflow—not whether every possible production cost is only $20.

Series roadmap

1. Prologue: the AI-agent experiment and operating model
2. Part 1: building a personal asset-management automation system
3. Part 2: producing and publishing YouTube Shorts
4. Part 3: producing multilingual webtoons and motion comics
5. Part 4: connecting the outputs into a multi-channel publishing system

The next article explains how repetitive spreadsheet work became a web dashboard and Telegram report, what remained manual, and what I learned by using the system in daily life.

Continue: Part 1 — Building a Personal Asset-Management Automation System

This article is an English localization of a production record first published in Korean. Korean original / 한국어 원문.