When you do the same thing for the fifth time, it’s time to turn it into a tool
Skip to main content

When you do the same thing for the fifth time, it’s time to turn it into a tool

Illustration: A man wearing glasses compares the open "The Analects" under a desk lamp, the Confucius question and answer game is on the laptop screen, and the street view of Taipei at dusk is outside the window

This article was originally published in “Vista Newsletter (opens in a new tab)” No.159

Hello, hello!

Yesterday was November 15th, the birthday of my third- and fourth-grade tutor, Teacher Cai.

A few days ago, at noon, Teacher Cai, I, and a former classmate went to the Grand Hotel to have Hong Kong-style tea. When I arrived at the restaurant, I realized that the teacher’s birthday was coming soon. My classmates were very considerate and even celebrated the teacher’s birthday early. The sky outside the window was gray, but I had a great time eating that meal.

Therefore, in this article, I would like to say to Teacher Cai here: Happy birthday.

Three days after eating that meal, I started my busy life again. I accompanied my mother to the household registration office, started working on the new book “AI Unmanned Company” for the four schools, went to Neihu for two consecutive mornings to provide AI tutoring for a business owner, guided a graduate student to deal with thesis problems, and had dinner and chatted with friends in the evening.

In the gaps between these trips, I did two things that were very interesting in retrospect: one was to consolidate work that had been repeated several times into a set of tools; the other was to design a class for a group of Chinese students to use AI to read “The Analects of Confucius”.

On the surface, these two things are unrelated, but when I reviewed it at night, I discovered that they are telling the same truth: AI makes it easier to do things, but it still depends on people to judge what is worth doing and whether the things produced are right.

For the fifth video, I condensed the process into a tool

From the beginning of November to now, I have made several teaching videos (opens in a new tab), such as: how to write an annual review, how to work backwards from the oral examination date, what stations to submit to journals, what are the four things to do first after being rejected, how to negotiate salary when you get a job offer…

Each one has the same format: a cartoon character stands on the left side of the screen and talks, the whiteboard on the right shows the key points step by step, the narration is synthesized by speech, and finally two versions, horizontal and vertical, are output, one for YouTube and one for Instagram Reels.

When I made my first video, everything was new. The characters had to be drawn, the layout had to be arranged, and the narration had to hear the wrong words sentence by sentence. It took a lot of time.

By the time I made the third video, I began to find myself repeating the same steps and falling into the same pitfalls.

By the fifth and sixth videos, I decided to stop and condense the entire process into one tool.

I use Claude Code’s skill function. You can think of a skill as a work manual for AI: it clearly states how many steps this type of work must be divided into, what tools to use at each step, and what areas must be paid attention to. Next time I just say “make another instructional video”, and the AI ​​will follow this manual and start a new project step by step.

The process of organizing the manual is actually to write down the pitfalls I have encountered during this period one by one. Let me give you a few examples:

  1. Speech synthesis will not change the tone. When we say “not” or “one”, “no” and “一” will naturally become two tones, but the speech synthesis will literally pronounce them into four tones, which sounds weird. So when writing a narration, try to avoid these words.

  2. Words with broken pronunciation will be mispronounced. For example, the word “行” in “calendar” may be pronounced as “行” by speech synthesis. The solution is to change the reading list to homophones, and the subtitles on the screen will still be displayed as the original text.

  3. The picture should change significantly every four to six seconds. Stop a scene for too long and the audience slips away. I wrote a check step in the manual. Any place where the picture has not been changed for more than six seconds will be listed, allowing me to decide whether to make up for it.

  4. Instagram Reels has a maximum length, and the narration should be controlled within approximately 780 words.

Behind each of these rules is a reworking. After writing a manual, you don’t have to step on the same pitfalls a second time.

The day the manual was written, I used it to make several new videos. In the future, the starting cost of each tablet will be much lower than before.

When should you turn something into a tool?

Many people ask me, where to start when using AI to improve efficiency?

My answer is: everyone’s answer is different. You can start with the thing you do the most times.

However, don’t act too early. When you do something for the first time, you still don’t know which steps are necessary and which pitfalls will appear again and again. If you rush to write down the process at this time, most of what you write will have to be completely destroyed and started over.

My own habit is to use three signals to judge:

  1. When you do it for the third time, start taking notes. You don’t need to write it completely, just write down “Where did you get stuck this time” and “How did you solve it last time”.

  2. When doing it for the fifth time, organize the notes into a process. By this time you already know which steps are the same every time and which things go wrong every time.

  3. When you find yourself telling the AI ​​the same thing repeatedly, it’s time to write a tool. If the same reminder is given more than three times, write it into the manual and let the AI ​​remember it.

This method is not only suitable for making videos. Reports to be written every week, accounts to be compiled every month, and letters from students to be answered every semester, as long as they are recurring tasks, can be handled this way.

Tools are not designed from the beginning, but grow from reworking them again and again.

Use AI to read “The Analects of Confucius”, AI missed a word

Another thing that took me a lot of effort this week was to prepare a Vibe Coding practical workshop for the “Analects” class of a certain university’s Chinese Department.

Students in this course have already read the growth history of Confucius, Confucius’ learning and teaching, and the portraits of Confucius’ disciples. In the last few weeks of the course, the teacher hopes that students will translate the classics they have read into digital works. My task was to spend three hours with my classmates, who had barely written any procedural formulas, and make each of them a playable mini-game of “The Analects of Confucius”.

The first step in preparing lessons is to do it myself.

I used Google’s Antigravity and asked it to play a question-and-answer game called “Teach students in accordance with their aptitude, guess what Confucius answered.” For the same question, Confucius will give different answers according to the personality of his disciples. Players have to guess what Confucius will say this time.

As a result, it was written in about a minute. One file, 906 lines of code, and I didn’t write a single line. You can play it after opening it, and the graphics are pretty decent.

At this point, it sounds like a story about powerful AI. But then, I did something that my classmates in the Chinese Department must be better at than me: checking the original text.

I compared the twenty quotations that appeared in the game, one by one, with the original text. Twelve of the paragraphs are completely consistent with the original text, and the rest are mostly vernacular explanations or differences in variant characters.

But there is one paragraph missing a word.

The original text of the chapter “Li Ren” is: “When parents give advice, they do not obey their wishes, but they are respectful but not disobedient, and they work hard without complaining.”

AI wrote: “No matter how much your parents give you advice, you will not obey your will, but you will respect it without disobeying it, and you will work hard without complaining.”

There is one less “and”.

The difference in one word is probably not something most people would notice when playing games. But for those who study Chinese, the difference of one word is the starting point of edition studies.

Where you are better than AI

This missing word later became a case that the entire brief returned to again and again.

When preparing lessons, I kept reminding myself: Don’t appease this group of students by treating them as programming laymen. They cannot write programs, but they have three abilities that AI cannot match: reading the original text, distinguishing authenticity, and paying attention to the source.

AI writes quickly, but who is to judge whether it is written correctly? The answer is this group of students.

So I wrote a sentence in the briefing: Ancient prose written by AI should be treated as homework submitted by students for correction.

The specific method is not difficult either. Ask the AI ​​to make a list of all the quotations in the game, and attach the title of the article; then use Wikisource or an annotation at hand to compare it word for word; and finally indicate the source and authorization of the original text at the end of the game page. There is no copyright issue in the original text itself, but there are modern vernacular translations and annotations, which should be clearly distinguished when citing.

There is another point to be careful about: games are interpretations, not rewriting classics. There is no problem in making Confucius a character in the game; but if you let him say something that was not said in the original text, you must clearly indicate that it is your interpretation.

The brief ended up being nearly two hundred pages long, plus an installation manual and a manual for students to study at home. It is one of the textbooks that I have spent the most time on this semester, because what I want my classmates to take away is not “AI is so powerful”, but “It turns out that the books I have read for so many years are so useful in the AI ​​era.”

The first sentence of the first chapter of “The Analects of Confucius”: “Learn and practice from time to time.” I wrote on the second page of the briefing: Today’s “practice” means doing it.

One-on-one, the most tiring and the most fulfilling

Finally, let’s talk about what takes up the most time this week: one-on-ones.

On Saturday and this morning, I went to Neihu to provide AI tutoring for a business owner. He is my long-term tutoring student. Every time we have class, we start from the real problems encountered in his company and think about where AI can help.

One-on-one lessons are the most rewarding of all my jobs. It can’t follow the briefing. It has to listen carefully every minute to what the other person is saying, where it is stuck, and then come up with the solution that best suits him on the spot.

But it’s also one of the most fulfilling. In a large classroom, it is difficult to be sure that everyone understands; when you are one-on-one, as soon as the other person’s eyes light up, you will know that it is done.

On the same day, I also guided a graduate student to deal with problems on the thesis. Her question was very specific. Together we took apart the stuck areas and found the way forward step by step.

During these three days, I spent most of my time giving one-on-one lessons and friends. There is a sense of accomplishment, but it is also very exhausting. A manual that was originally scheduled to be written remained untouched for several days.

I reminded myself during the review: One-to-one value is very high, but it is relatively difficult to replicate. That’s why I want to condense things I’ve done many times into tools, and make videos and manuals out of what I’ve said many times. Only in this way can the experience gained in one-on-one sessions be used by more people.

Three things you can take away from this article

Let’s summarize the key points of this article:

  1. When you do the same thing for the third time, start taking notes; when you do it for the fifth time, organize it into a process; if you find yourself telling the AI the same thing repeatedly, write it down as a tool.

  2. AI makes it easier to do things, but it still depends on people to judge whether it is right or not. The expertise you have accumulated in this industry is exactly what is needed most in the AI ​​era.

  3. Please regard the content written by AI as homework submitted by students for correction. Especially the citations, figures and sources must be checked back to the original text.

Recent events

The new book “AI Unmanned Company” is undergoing the final proofreading

My new book “AI Unmanned Company” is expected to be published by Booker Culture in December. It is currently being published in four editions these days. This book is about: There are no employees in the company, just you, plus an AI team designed by you and directed by you. This article talks about “turning repetitive tasks into tools”, and there is a more complete approach in the book.

There is a five-question quiz on the new book page, and you can also sign up for release notifications, which will notify you as soon as the book comes out.

View the new book page of “AI Unmanned Company”

Seven-day challenge camp, registration deadline is November 28

The seven-day challenge camp starts on November 30th and registration closes on November 28th. One task letter a day, and in seven days you can make a small tool that you can use every day. You can put it on the Internet and get the URL without having to write a program. The Chinese students mentioned in this article can make a small game in three hours, and you can too.

See the introduction of the seven-day challenge camp (opens in a new tab)

December 20, Academic Research and Writing Workshop, early bird until November 29

If you are a graduate student or university teacher, on the morning of December 20 (Sunday), I will have a three-hour practical class in Taipei to teach you to use AI Agent as a research co-pilot to read faster, make better corrections, and at the same time maintain academic ethics. This article talks about checking citations and sources, and there will be a complete exercise in the workshop. Sign up before November 29th to enjoy the early bird price.

See the complete introduction and registration of the workshop

Haven’t you got the “Letter to Your Supervisor” sample set yet?

A collection of free templates released a while ago contains 18 templates for writing letters to your advisor, ranging from asking about your willingness to be mentored for the first time to thank you letters after graduation. If you have classmates who are writing papers, you are welcome to forward them to them.

Free download of the “Letter to the Supervising Professor” sample collection

Finally, I want to hear your story

Do you have a task that you do repeatedly every week or every month? If you could hand it over to AI, which one would you most want to hand over first?

Feel free to leave a message and tell me. See you next time!

Vista

Next step

Bring this topic to your team

Vista Cheng speaks and runs workshops on AI workflows, writing and content systems for companies, schools and public agencies.