7 AI Utilization Techniques from Former Goldman Sachs MD Kei Tanaka

Former Goldman Sachs. Just hearing this title makes someone seem like they're from another world.

But after learning how this person uses AI, that impression was completely overturned.

Kei Tanaka.

He graduated as the top student from the Department of Physics, Faculty of Science and Technology at Sophia University. He passed 53 interviews to join Goldman Sachs in 2007. He was involved in over 500 investment projects, executing more than 60. The investment scale he managed was approximately 400 billion yen, exceeding 1.2 trillion yen on an enterprise value basis. Ultimately, he served as Managing Director (executive level) and Co-Head of the Investment Banking Division in Japan. He spent 17 years at the firm.

Looking at his career alone, he is a "person above the clouds."

Mr. Tanaka said in an interview:

Research and thick document preparation that used to take 3 weeks can now be finished in about 1 hour when combined with AI.

3 weeks to 1 hour.

The scale is crazy. It sounds like an exaggeration. But it's a story he actually tells.

Hearing this, many people think, "He's a former elite, his brain is just wired differently."

But something caught my eye.

When you actually follow Mr. Tanaka's AI usage, it's surprisingly simple. What he's doing isn't about talent or secret tricks. It's about a humble "framework" that anyone can imitate starting today.

It's long, so I recommend saving it if you want to look back later.

Why Most People Stop at "Convenient Tools"

First, where is the difference between Mr. Tanaka and ordinary people? If you grasp this, all 7 methods will connect.

Most people open AI as a "smart search." They throw a question. They get an answer. End of story. It's convenient. But that's a way anyone can use it. AI becomes just an information vending machine.

That's not what Mr. Tanaka sees in AI. From research to hypothesis building, decision support, and automatic document creation. He leaves this entire sequence to AI as a single flow. He doesn't place AI as a "tool to rely on occasionally," but as a "partner that reconfigures the very foundation of work."

The difference between those who use it instead of searching and those who use it as a device to speed up judgment is like heaven and earth.

I see two things that run through his usage. One is "Templatization." Putting the judgment process in your head into words and transferring it to AI. The other is "Compression." Producing the same quality in a fraction of the time. Keep these two axes in mind as you look at the seven methods.

Method ① Pass the "Template" for Judgment to AI First

The first one is the most effective. And it's a move almost no one does.

When you're lost at work, do you worry from scratch every time? "Should I proceed with this plan?" "Should I release this product?" Every time, your head gets messy. It's common, right?

Mr. Tanaka is different. He decides on the "template" used for investment decision-making beforehand and has the AI memorize it. By just pouring in the collected information and his own hypotheses, he creates a state where a draft for judgment comes back. It's fast because he doesn't think from scratch every time.

This isn't just for investors. The more judgments you repeat, the more you can turn them into templates. Put it into words once and give it to the AI. First, tell it this:

1Please help me with my decision-making from now on. I will give you the "template" for judgment first. From now on, when I throw a project at you, please always evaluate it based on these four points.
2① Demand: Are there people who really want it? (Specifically who?)
3② Reason for winning: Why does it make sense for me to do it more than others?
4③ Worst-case scenario: What do I lose if I fail?
5④ First step: What is the minimum action I can take this week?
6Please keep each item within 3 lines.

Then, just pour in the plan you're hesitating about. Demand, reason for winning, worst-case scenario, and the first step will come back filled in.

What I do there is add a correction: "Make the first step in ④ even smaller. Make it a granularity that can be done in 30 minutes today." This is where the judgment finally moves forward.

Let AI go from 0 to 1, and humans go from 1 to 10. Don't reverse the order.

People who don't have this template go back to the starting line every time and repeat the same worries for a lifetime. You should be different.

Method ② For Research, "Prepare the Night Before"

This is the identity of the "3 weeks → 1 hour" mentioned at the beginning.

Mr. Tanaka says he runs the AI's Deep Research (a function where the AI searches the internet itself and summarizes it into a report) the night before. The next day, he spends 1 to 1.5 hours during bath time to input it all at once. He creates a state where the preliminary research is finished while he is sleeping.

Most people start searching "on the spot" when it becomes necessary. That's why time melts away in research. Mr. Tanaka prepares research as "something to receive later." This is the decisive difference.

You can do it from today. Just throw this before you go to bed at night.

1Please summarize the following theme using Deep Research by tomorrow morning.
2Theme: [Topic]
3What I want to know is: ① Facts and figures known now ② Arguments for and against ③ Links to primary information that served as the basis.
4You don't have to draw a conclusion. Just list the materials for judgment in bullet points.

In the morning, look at the results while drinking coffee. Just with that, your day starts from a state where "research is already finished."

Method ③ Talk by Voice and Reduce Writing Time to 1/10

Once you speed up judgment with templates, next you cut the time spent "writing."

Mr. Tanaka makes full use of AI transcription for document creation. He speaks ideas for about 5 pages of A4 paper orally for about 2 minutes. The AI instantly turns that into text and organizes it. He says this reduces the time spent typing sentences to about 1/10.

The point is not to stop thinking. Don't try to match the speed of your head to the speed of typing on a keyboard. Think while talking, and leave the organizing to the AI.

Pass the messy monologue blown in by smartphone voice input like this:

1Below is a memo before organization that I spoke via voice input. Since it's in spoken language, please keep the meaning unchanged, just organize the order, and summarize it into 3 key points and short sentences explaining them. Please do not erase my phrasing or verbal habits.

The sentence "Please do not erase my verbal habits" is effective. Without this, the AI overwrites it with its own uniform writing style, resulting in that "AI-like text."

To be honest, I also blew in the framework of this article by voice while walking. Let the AI take over the heaviest first step of "Okay, let's write." Just that drastically lowers the hurdle to starting.

Method ④ Write Instructions "As if Teaching a Middle Schooler"

Even if you know how to use the tools, if the instructions are sloppy, what comes out will be sloppy. Mr. Tanaka's words are directly effective here.

He says he gives specific and clear instructions, even for complex things, as if teaching a middle or high school student. Instead of "do it nicely," write the premises and conditions assuming the other party knows nothing.

The difference appears in this one point.

• ✗ "How to start a blog as a side hustle?"
• ◯ "An office worker wants to start a blog as a side hustle. Available time is 1 hour a day on weekdays. The genre hasn't been decided yet. Under these conditions, I want you to decide what to do in the first month together, with priorities."

Even with the same AI, what comes back is completely different. Think of the other party as a smart but uninformed junior, and pass the background. Just that will make the accuracy jump.

If it's a hassle, it's easy to add this sentence to the beginning of the instruction.

1To perform this task with high accuracy, please ask me 5 questions first that you want to check with me. Please start the task after I answer.

Let the AI ask questions first. Then, the premises you forgot to pass will be naturally filled.

Method ⑤ "Possess" Role Models into AI

Once you improve the accuracy of instructions, next you increase perspectives. This is a way to exceed the limits of thinking alone with AI.

Mr. Tanaka has the AI memorize the ways of thinking and expression styles of role models in various fields and has it analyze from those perspectives multi-dimensionally. Get opinions from angles that your own head alone cannot reach.

The weakness of solo brainstorming is that the field of vision does not go outside of yourself. You fill that by having the AI possess another personality.

1Please comment on my project from the positions of 3 people in order.
2① A skeptical investor (poke all the sweet premises)
3② A lazy reader (would they feel like continuing to read?)
4③ Myself one year from now (will this remain as an asset?)
5For each, give 1 good point and only 1 harsh criticism.

It becomes a state where you are running a meeting with multiple people by yourself. To the returned criticism, you further layer: "To convince the reader in ②, how would you change the first 3 lines?" By repeating this back and forth a few times, the holes in the project are surprisingly filled.

However, do not swallow the AI's opinion whole. It is strictly a device for identifying issues. You hold the final judgment.

Method ⑥ Don't Let One AI Do Everything

This can be seen by looking at the tools Mr. Tanaka uses.

He lists ChatGPT, Claude, Perplexity, and Notta for voice input. In other words, he doesn't leave everything to one AI. He uses different AIs for different types of work.

Most people always ask the same one AI everything. But each has its specialty. Just by sorting them like this, the quality changes.

• Research and primary information collection → Perplexity (source links are attached to answers)
• Sparring and organizing long sentences → Claude or ChatGPT
• Voice transcription → Notta or Claude

If you try to do everything with one, everything becomes halfway. People who switch based on usage go ahead in accuracy by that much.

You don't have to get everything at once. First, add one research AI to your usual AI. That's enough to start.

Method ⑦ "Contentize" Yourself and Feed It to AI

The last part is about raising the accuracy of the AI itself specifically for you.

In interviews, Mr. Tanaka repeatedly emphasizes the "importance of keeping records." In fact, he even says it was a "waste" that he hadn't recorded the past. Keep a record of your thoughts, judgments, and words, and make them ready to be passed to AI. In other words, make yourself into content.

This is humble but effective. AI knows nothing about you. That's why it returns answers like an average stranger every time. But if you pass the log of your thoughts, the AI will start returning answers closer to you.

You don't need a large-scale diary. 3 lines a day is fine. Once a certain amount has accumulated, pass it like this:

1Below are notes of things I've thought about recently. From here, please extract my values, judgment habits, and frequently used words, and summarize them in bullet points. In the future, when writing sentences for me, please reflect these characteristics.

Once you do this, subsequent AI outputs will become "personal." For me too, just by passing one memo summarizing the tone of my communication, the corrections to the returned sentences have decreased significantly.

This is what I call "Knowledge is King." Before polishing prompts, organize the knowledge you pass to AI. This is the main keep.

To Those Who Thought "It's Because He's Elite"

After reading this far, some people must have felt this: "He's former Goldman, his natural intelligence is different."

It's the opposite.

These 7 methods don't require special qualifications or expensive tools. Turn judgment into a template. Prepare research the night before. Write by voice. Instruct as if teaching a middle schooler. Possess another perspective. Use different AIs. Record yourself. All are actions you can start today with free AI.

In the first place, Mr. Tanaka himself is a person who built up from the origin of failing middle school entrance exams. The reason his AI usage is superior isn't because he has talent. It's because he is calmly and thoroughly implementing the principle that anyone can imitate: "Turn the inside of your head into a template and transfer it to AI."

What's needed isn't technology. It's just a design that doesn't let the AI get lost.

Summary

As long as you open AI as a "convenient research partner," you haven't even drawn out half of its power.

• ① Pass the "template" for judgment first, then pour in the plan.
• ② Prepare research the night before.
• ③ Talk by voice and reduce writing time to 1/10.
• ④ Instruct specifically from premises, as if teaching a middle schooler.
• ⑤ Possess role models and run a multi-perspective meeting by yourself.
• ⑥ Don't let one AI do everything; use them differently by purpose.
• ⑦ Record your thoughts and make the AI dedicated to you.

If you end with "Oh, I see" after reading, tomorrow you will again hold everything manually as usual. Most people do that.

But you should be different. Try passing one of your usual judgments to AI today. That will be the first template.

Finally

What did you think?

The idea of "turning judgment into a 'template' and transferring it to AI" that we talked about today. This is something I myself run every day at the business site. Once you start running with this way of thinking, you can't go back to the original way.

And for a limited time only. I am distributing "12 Major AI Utilization Benefits" for free.

From the basics of AI utilization to a system that leaves 80% of work to AI. E-books, introduction guides, glossaries, and free individual consultations. I've packed all my current knowledge into these 12 items.

First, there are 6 e-books that create the "foundation" of AI utilization. If you read them in order from Volume 1, you will naturally acquire essential AI utilization.

• VOL.1 Prompt x Context Complete Mastery (The two major elements that determine the quality of AI output in 5 chapters. Start here.)
• VOL.2 Working AI Complete Mastery (The entrance to "working AI" called Claude Code and Codex.)
• VOL.3 Knowledge Complete Mastery (How to store and grow "information assets" that AI comes to read repeatedly.)
• VOL.4 Debugging Complete Mastery (Practical techniques to change "something is wrong" to "this is how it's different.")
• VOL.5 How to Leave 80% of Work to AI (From inventorying tasks to API/MCP integration, expanding the range of delegation at once.)
• VOL.6 Skills Complete Mastery (Creating a system where work ends with a single command. The final volume of the series.)

Next, there are 2 introduction guides you can start today. You can proceed without hesitation until you put "working AI" into your own computer.

• Claude Code Introduction Guide (For Claude users. From installation to initial settings.)
• Codex Introduction Guide (For ChatGPT users. From installation to initial settings.)

Furthermore, there are 3 pages and 60 words of AI terminology explanations that you can look up. All the words you "can't ask now" are here.

• AI Terminology Explanation ① | Basic Edition (Generative AI, Prompt, Hallucination—20 words about AI itself.)
• AI Terminology Explanation ② | Claude Code/Codex Edition (Terminal, Skill, Path—20 words for screens you touch every day.)
• AI Terminology Explanation ③ | Integration/Tools Edition (API, MCP, .env—20 words you hit in automation.)

And the 12th one.

• Free Individual Consultation (1 ON 1 / X GROWTH). I will tell you one-on-one how to grow your X communication in the shortest and fastest way and also gain advertising revenue. It's a reproducible method even for absolute AI beginners, so if you "don't know where to start," please come. Slots are limited, so please hurry.

A total of 12 items. And they are all free. Even if you don't participate in seminars or free individual consultations, you can get all the e-books, introduction guides, and glossaries.

It sounds like a lie, doesn't it? But it's true.

How to receive them is simple. It starts from joining the LINE Open Chat below.

Also, I will disclose useful information that cannot be shared on X in the Open Chat.

Click here to start.

I'll be honest. These 12 items weren't completed in one go. I remade them many times and pared them down to a form that even AI beginners can use without hesitation.

Let me say it just one more time. What's needed isn't technology. It's just a design that doesn't let the AI get lost.

Why don't you end the exhaustion of only seeking "answers" from AI today?

References

Mr. Kei Tanaka's statements and career are based on the following public information.

• The Keyperson Interview (Specific methods of AI utilization, "3 weeks → 1 hour", tools used)
• PIVOT "5 Rules of Former Goldman Sachs Kei Tanaka"
• Pen Online (Deep Research the night before, input during bathing)
• Wikipedia "Kei Tanaka" (Sophia University, top of department / 53 interviews / investment scale approx. 400 billion yen, enterprise value over 1.2 trillion yen / MD, Co-Head of Investment Banking in Japan / 17 years of service)
• Book: "People up to 100 Million, People from 100 Million: 'Trillionaire' Mindset Revealed by an Investor with 17 Years at Goldman Sachs" (Tokuma Shoten)