Have AI Build Automation That Runs in Production

Automated trading, video posting and ebooks, run unattended: what I decided, what I checked, and how they broke
What you'll learn

---

Write, in one line each, the three things a human decides when AI builds it (requirements, limits, stop conditions) for your own problem
Tell "it works" from "it does not break", and ask AI for the tests, the monitoring and the self repair that close the gap
Follow the design of three automations that really ran (trading, video posting, ebooks) and assemble your own version
Build inspection items in advance from real failures (an order that did not exist, a day lost to one line of configuration, the 25% trap)
Know the lines on rules and responsibility (investment advice, distance selling law, AI disclosure) and get to a form you can publish
Requirements

---

You have asked a generative AI (ChatGPT, Claude and so on) to do something for you.
No programming knowledge is needed. No equipment is needed (the handouts are text files).
Description

---

This course contains the use of artificial intelligence. The narration is a synthetic voice, the slides come from the same generator the instructor's own system uses, and AI was used to draft parts of the script. The script was written from the instructor's own operating records, and every fact and number was checked against those records.
A humanities student instructed a generative AI, in the summer of 2026, to build an operating system that runs unattended, and then ran it on a real account with his own money. He has not written a single line of the code. The only things he decided were three: the requirements, the limits and the stop conditions.
The course covers three automations that really ran. Automated trading in crypto assets (on a real account from August 20, 2026 to September 6, 2026, and on paper with the same setup ever since), a pipeline that turns articles into videos and posts them to YouTube and social media unattended, and a pipeline that builds and inspects ebooks. More time goes to the stories of them breaking than to the designs. The 246 yen of an order that did not exist, the day erased by one line of configuration, the 25% gain the AI refused, the incident where the tests fired notifications at the real channel. Every number comes from the instructor's own records.
There are 9 handouts to take away. The original text of the first instruction and the questions the AI returned, a template for requirements, limits and stop conditions, the public version of the risk policy, the pre production checklist, the audit prompt and the two critical findings it produced, the machine inspection items for videos, the build rules for ebooks, the incident timeline, and the worksheet. All of them come from the records of the instructor's own system.
This course is not investment advice. It does not recommend any exchange or any product, and it does not promise a return. Over the period it ran on a real account, the instructor's account did worse than simply buying a fixed amount on a fixed schedule. What is taught is what the person who hands the implementation over decides, checks, and watches break.
Each lecture is one point, 1 to 4 minutes. The instructor's note at the end of each lecture is where the instructor, not the AI, says what he decided at the time. Questions are answered by the instructor in the Q and A.
Narration: Kokoro-82M (Apache-2.0), a synthetic voice, run locally. The Japanese edition of this course is narrated by VOICEVOX: Kurono Takehiro (CV: Garo); that credit applies to the Japanese edition only.
Who this course is for

---

People who tried having a generative AI build something and stopped because it does not run, it stops, or it breaks.
People who want to run their own work unattended but do not know what to decide before handing it over.
Homepage

---