In the AI Era, Only Those Who Turn Their Philosophy into External Memory Will Win

People often say, "I had AI write this."

I recently wrote a long-form article. It was read quite a bit, and I was asked several times, "Did you write that with AI too?"

Every time, I felt a sense of discomfort. I didn't make it write; I entrusted it.

It's only a slight difference in phrasing, but the meaning is completely different. Whether or not you can understand this distinction will determine how you use AI from here on out.

Right now, I use AI as a partner for almost everything: X (Twitter), long-form articles, client work, operations, and business strategy. The foundation for entrusting these tasks is my "Philosophy and Thought Database" (hereafter, Philosophy DB). It's an external memory of my thoughts managed in Notion.

In the coming AI era, only those who turn their philosophy into external memory will win. Regardless of which AI is at the front end, the output of a human with external memory maintains its uniqueness. The output of those without it will be swallowed by the average, even if they use the latest models.

In this article, I will write everything about the contents, how to create, how to update, and how to use the Philosophy DB.

For those who read to the end, I'm giving away the structure template of the Philosophy DB I use. It's designed so that when you load it into Claude, it will guide you through designing your own dedicated Philosophy DB. I'll be distributing it to those who repost, so please feel free to receive it.

"Making it write" and "Entrusting it" are different things

It's easier to understand if you think about it in terms of an organization.

There's a difference between "dumping" work on a subordinate and "delegating authority." Dumping is throwing work at someone without sharing the criteria for judgment. The resulting product belongs to no one and often misses the mark. Delegating authority is entrusting work to someone who shares your criteria for judgment. The resulting product reflects the will of the person who entrusted it. It stands as a work of the person who delegated it.

It's exactly the same for AI.

"Making AI write" is dumping. You put "write about XX" into a prompt and copy-paste what comes out. The resulting text will be roughly the same whether it's ChatGPT, Gemini, or Claude, and no matter who does it. This is because the judgment criteria only exist in the "average within the AI."

"Entrusting AI" is delegating authority. You let the AI refer to your own philosophy and then entrust it with execution. The resulting text is indistinguishable from what you would have written yourself. This is because the judgment criteria are within you, and the AI is reading them.

Most "AI writing" in the world is the former. That's why it's thin. It's text where no one would notice if you changed the signature. That is the true nature of text made by "making AI write."

The shallowness of the term "Personal AI"

More people are saying they want to "entrust" rather than just "make it write." I've been seeing terms like "Personal AI" or "Persona AI" a lot lately.

When you look at the contents, they're usually doing things like this:

• Writing "You have a XX personality" in the system prompt
• Making it mimic their tone or sentence endings
• Shoving past tweets into training data
• Creating detailed character settings

These are all "imitation of style," not "inheritance of philosophy."

Recently, explanatory entries for setting methods like "SOUL.md" have been circulating. When I read them, they mostly focus on the AI's tone, endings, personality, preferences, and persona settings, and often don't delve into "how to give it a philosophy." It's an approach of tweaking expression to create a sense of "self-likeness." I don't use it that way. Even if you're happy that it "became like you" by setting expressions, your own philosophy isn't reflected in the core of the output, so it ends up being just momentary satisfaction.

Style is clothing. You can take it off, change it, and it varies by day.

Philosophy is the skeleton. You can't take it off, you can't change it, and it stands the person up from the inside.

Even if you mimic the clothes, you don't become that person. You only become that person when you capture the skeleton. Setting a "style" for AI is just changing its clothes.

Most people who talk about "Personal AI" haven't verbalized their own philosophy. The order is reversed.

Furthermore, the very perception of AI utilization is different. Many people think that "throwing tasks at AI" is AI utilization. It's not. AI utilization is "designing the entire operation and placing AI and humans within it." You separate and place the parts for AI to handle (solving/executing) and the parts for humans to hold (questioning/choosing/having will). This design requires the judgment criteria of your own philosophy. People without judgment criteria cannot design the placement, so they end up just dumping everything on AI.

Before entrusting AI, structure what you are thinking. Start from there.

Giving it a philosophy means putting your philosophy outside

So, what does it specifically mean to structure your philosophy?

You might have an image of "giving" a philosophy as installing something inside the AI. But the reality is the opposite.

You put your philosophy outside. You structure it and place it outside yourself in a referable form.

The AI reads it every time. After reading it, it applies the philosophy to the task at hand to make judgments. This is what "entrusting" is all about.

In other words, "giving AI a philosophy" isn't about doing something to the AI. The core is verbalizing your own philosophy and putting it outside. The AI is merely a device that refers to it.

Many people misunderstand this. They think that if they tinker with the AI, a philosophy will dwell within it. No. Unless you verbalize it yourself, nothing will dwell in the AI.

Therefore, the main topic of this article isn't how to use AI. It's about "how to put your philosophy outside, how to structure it, and how to update it." If you want to entrust AI, you have to start by putting your philosophy outside.

One day, I couldn't distinguish my writing from AI's

You might think, "Is it worth going that far?" Let me share a bit of a personal story.

Since my student days, I've run local media and several blogs. This might sound like bragging, but my standard score (Dev) for modern Japanese was over 70. Writing has been part of my livelihood for nearly 20 years. I have a habit of scrutinizing Japanese at the level of particles, punctuation, and character count. Since I make a living from content marketing, I'm aware that I'm more particular about writing quality than most.

I read a text that I had the AI write by entrusting it and letting it refer to my Philosophy DB, side-by-side with a text I wrote myself.

I couldn't tell the difference.

The rhythm of the sentences, the choice of vocabulary, the placement of punctuation, the progression of the argument, where to emphasize and where to omit. All the "judgments I would have made" were right there. For a moment, I was seriously confused: "Did I write this?"

It's only natural. The Philosophy DB contains not just philosophy, but also regulations for written expression (phrases to avoid, preferred sentence endings, punctuation rhythm, the ratio of Kanji to Hiragana). The AI reads that and produces text after performing all the scrutiny I would have done.

This is when I realized. Putting your philosophy outside means putting out the rules for how that philosophy manifests as writing along with it. If you go that far, the AI can write my text in my place.

And when you reach this point, only two jobs remain for humans: deciding what to write and deciding how to write. Both are manifestations of will. AI can write anything, but it cannot decide "what should be written." AI can build any structure, but it cannot judge "whether that structure connects my question to the goal." Will remains in the realm that AI cannot touch. Therefore, humans who entrust AI can concentrate on their will.

Contents of the Philosophy DB

The device that creates that destination is the Philosophy DB. From here on, it's about implementation.

I manage the entire Philosophy DB in Notion. The structure has three layers.

1. General Theory Page

A page that summarizes the overall picture of your philosophy in one sheet. When entrusting AI, you have it refer to this page first.

My general theory consists of four blocks.

[Block 1: Background and Formative Experiences] The background of how the philosophy was born. In my case, things like this are written:

• I grew up around the family business and breathed in commerce and management like air.
• I studied accounting in university. My eye for looking at organizations through numbers was formed here.
• I joined a consulting firm as a new graduate and quit in one year. I felt uncomfortable with "advice that doesn't take responsibility for people's lives."
• I started my own company and switched to a style of running long-term alongside the people I work with.

The point of writing your background is to give the AI the context of where the philosophy came from. Even with the same "view of work," the content differs between someone who saw a family business and someone who grew up in a salaryman household. If you don't write your origins, the philosophy floats in mid-air.

Without a background, the AI will write "text of an average consultant." Uniqueness disappears, and it becomes text that anyone could have written. This is the most common failure pattern when entrusting AI.

I also write "people and works that influenced me" here. In my case, Erich Fromm, Aristotle, Kazuto Ataka, and Mr. Children are listed. Whose words you've been immersed in is the very skeleton of your philosophy. If you don't write this, the AI will arbitrarily supplement with "safe references as general knowledge." That's the cause of creating text that is furthest from yourself.

[Block 2: Characteristic Data] Objective data on your cognitive characteristics and behavioral tendencies.

• Top 5 StrengthsFinder: Individualization, Achiever, Command, Strategic, Activator
• MBTI: ENTJ
• Thinking habits: Abstraction runs ahead, concrete details are filled in later
• Behavioral habits: Quick judgment, can't wait, hands-on before delegating to others

Writing this allows the AI to match the granularity of its suggestions. It will return suggestions based on your characteristics, like "Since you have high Activator, a style of thinking while running suits you."

Without characteristic data, the AI will make "rounded suggestions for everyone." It will return advice that is correct but doesn't suit you, like "First, let's make a plan and build consensus with stakeholders before starting." It becomes the text of a person who can't move.

[Block 3: Pillars of Core Philosophy] The pillars of values and worldviews you hold dear.

• Work is a means of self-actualization. It's not a means to earn living expenses.
• The starting point is always a sense of discomfort with modern society. Start by questioning what is considered normal.
• Don't think of technology and human consciousness separately. Both evolve inseparably.
• Prioritize meaning over happiness. Happiness is a state; meaning is a direction. Happiness without direction collapses quickly.
• Look at human evolution on a long time axis. Thinking in 10-year units changes today's decisions.

This is the very core of the Philosophy DB. When the AI makes a judgment, it ultimately refers here.

Without core philosophy, the AI only returns "greatest common denominator truths." Text that no one can disagree with, but no one gets excited about. This is the true nature of the "AI articles" overflowing in the world.

[Block 4: Practical Logic] A collection of rules on how to translate core philosophy into real-world judgments.

• "Individualization" is the theme of the era. Individualization is democratized by AI.
• Therefore, create services optimized for each individual rather than products for everyone.
• However, only humans can update physical existence and philosophical worldviews. Do not outsource this.
• Default to judgments that prioritize long-term meaning over short-term efficiency.

If you just have a philosophy, you won't know "so, how do I move?" when it's time to make a judgment. Writing down practical logic directly connects philosophy to action. When entrusting judgment to AI, the output is completely different depending on whether this exists.

Without practical logic, the AI's output will be like "someone who says grand things but can't act." It can talk about philosophy, but it won't say what to do tomorrow. It's text that feels good to read but leaves nothing behind. This is the most wasteful failure pattern.

By the way, I also write the "regulations for written expression" in this block. Phrases to avoid, preferred sentence endings, punctuation rhythm, the ratio of Kanji to Hiragana, and citation etiquette. A set of rules for putting philosophy into writing. When you write this much, the AI's output becomes your own at the level of expression. The phenomenon of "indistinguishable from my own writing" mentioned at the beginning happens because I write this much. The more you write, the more the AI becomes you.

These four blocks are the general theory. It fits on one page, but each block has considerable depth. My general theory is currently about 20 manuscript pages long.

2. Individual Databases

A collection of concrete databases that support the general theory. Philosophy DBs divided by theme, such as "View of Work," "View of Family," "View of Technology," "View of Life and Death," and "View of Organization."

The general theory alone can be too abstract, and the AI might not be able to break it down into specific tasks. In such cases, I have it refer to the individual databases as well. For example, if I'm writing about "work" in an X post, I have it read the general theory plus the "View of Work" individual database as a set.

Individual databases are about 10 to 20 pages. I create a new page every time a theme increases.

3. Update History Page

A page to record as a pair whenever the general theory is updated. When, what, and why it was changed.

Without this, philosophy changes "before you know it." That's not philosophy; it's just a flow of mood. Keeping an update history allows you to see the changes in your philosophy chronologically. You can see the difference from yourself six months ago.

I'll distribute this three-layer Notion template at the end of the article.

How I update it

So far, that's the talk about structure. But just creating the structure will make the Philosophy DB die. If you don't keep updating it through daily operation, it will quickly become a relic of the past.

The flow of updating is in two stages. First, collect materials. Next, dig with AI. This order is important. If you try to start from AI, nothing will come out. If you ask AI to "extract my philosophy" without any materials, the AI can only return generalities because it has nothing.

Follow the order. Materials first, AI later.

Collecting materials: Chat tools, recorded data, honest talk

Philosophy doesn't come out when you try to sit down and write it. Conversely, it leaks out in places where you aren't prepared. So, the first step is to gather places where the "unprepared self" remains.

I use four types of materials:

1. Private chat tools

LINE or personal DMs with people you feel comfortable with are a treasure trove of true feelings. I export the recent logs once a week. Interactions with friends, family, and old acquaintances are particularly effective, rather than work partners. Because it's a place where you talk with zero pretense, philosophical tendencies you haven't even noticed yourself are coming out. It doesn't have to be LINE; Messenger, Discord, personal DMs—anything is fine.

2. Speech logs from work chat tools

I use Slack with my work team. Messages I send to subordinates or colleagues, especially those conveying judgments, are clusters of philosophy. I pick up posts where I've written "I'll do this," "I won't do this," and "for this reason" as logs once a month. The accumulation of daily judgments most accurately represents your judgment criteria. It's the same whether you use Slack, Teams, or Chatwork.

3. Honest talk with people

There are moments when talking to someone I trust where I feel, "Oh, I just said something important." If I can record it on the spot, that's best, but if not, I leave a monologue of "what came up in that conversation earlier" within the same day. This has the highest density as material.

4. Recorded data

I record things I think of while moving or walking as audio. I don't talk to someone; I talk as a monologue. When I talk, words come out faster than when I write. I get prepared when I try to write, but it comes out when I talk. Notta, voice memos—anything is fine. In the same way, recorded data from Zoom or interviews, and logs of voice messages also become materials.

I keep accumulating these materials in a folder like "Philosophy Materials." I can organize them later. Just keep accumulating.

Digging with AI: Letting Claude read and extract

Once the materials are accumulated, the next step is to let the AI dig through them.

This is my method. I give the materials to Claude and hit it with a prompt like this:

Please read these utterances and texts and extract the parts related to my philosophy. I want you to extract the following:

- Persistent obsessions that appear repeatedly - Objects of strong discomfort - Criteria used during judgment - Perspectives that deviate from generalities

Tell me the extracted results along with which part of my general theory page structure (Background, Characteristics, Core Philosophy, Practical Logic) they fit into.

When I hit it with this, Claude returns output like, "You have recently been making judgments based on the criteria of XX regarding OO. This needs to be judged whether to be added to XX in the 'Pillars of Core Philosophy' or established as a new pillar."

AI structures and shows you tendencies that you wouldn't notice just by reading them yourself. This is effective.

Writing back: Adding to General Theory and Individual Databases, recording in Update History

You write back what came out from digging with AI into the general theory or individual databases. If you don't write it back, it will be buried in the dialogue and never referred to again.

What I always do when writing back:

1. Add to the relevant block (Background / Characteristics / Core Philosophy / Practical Logic)
2. Record "when, what, why, and the trigger" in three lines on the Update History page

The update history only needs to be three lines. What was changed, why it was changed, and what the trigger was. If you leave just this, you can trace the transition of your philosophy later.

Always move the general theory and update history as a pair. I've made this an absolute rule for myself.

Bonus: Materializing the AI dialogue itself

When having deep conversations with Claude, there are moments when your own philosophy deepens on its own within the dialogue. While answering questions Claude throws at you, things you couldn't verbalize until then become words.

This dialogue itself also becomes material. If you ask "Summarize the parts related to my philosophy that came out in this dialogue" at the end of the conversation, Claude will extract them. Then you write them back again.

How I actually entrust it

The structure is made, and the mechanism for updating has started to turn. From here, it's about the exit. All that's left is to use it. I'll write three ways I entrust tasks.

X Post Operation: Putting out fragments of philosophy

Starting with the lightest use. I divide X post operations among five Claude agents.

• Agent 1: Picks up seeds for today's post theme from the Philosophy DB
• Agent 2: Expands the seeds into post drafts
• Agent 3: Refines the style
• Agent 4: Adjusts the character count
• Agent 5: Final check

All five refer to the same Philosophy DB. Because the judgment criteria are shared, all the resulting post drafts become "fragments of my philosophy."

I only do one thing. I choose from the resulting post drafts, make fine adjustments if necessary, and post. That's it.

If the Philosophy DB didn't exist, the five agents would start running in different directions. The resulting post drafts would be a collection of generalities unrelated to my philosophy. People who say "AI posts are thin" are usually in this state.

Writing Long-form Articles: Systematically expanding philosophy

Next is something heavier. For long-form articles, I only decide the theme and direction and have Claude Code write it.

The specific steps are as follows:

1. Tell Claude the theme and 3 to 5 points I want to write about.
2. Have it refer to the relevant pages of the Philosophy DB (General Theory + related Individual Databases).
3. Instruct it: "Write for this audience, at this length, based on this philosophy."
4. Read the first draft and point out where the flow of the argument is weak.
5. Incorporate related concrete episodes (important).
6. Repeat corrections 3 to 4 times.
7. Finally, adjust the sentence endings myself to complete it.

The point is to have it refer to the Philosophy DB first. Whether you do this or not makes the quality of the first draft about two levels different. If you don't have it refer, the resulting text will be a mass of generalities common on the internet. If you have it refer, it becomes text written with my judgment criteria.

It's not just that "the quality of the writing improves." The subject of judgment changes from the AI to yourself. This is what "entrusting" is all about.

Business Strategy: Launching the future from philosophy

Finally, the deepest use.

I'm working with Claude to flesh out the concept for a new business I recently launched. The method is like this:

• Have it refer to the entire Philosophy DB.
• Have it think together: "What kind of business should a person with this philosophy launch in this era?"
• Scrutinize the resulting ideas one by one: "Does this align with the philosophy?" "Is there any discomfort with this?"

When I do this, the ideas coming from the AI align with my philosophy from the start. No noise like "a business that seems profitable but doesn't suit me" comes out.

Business strategy is a task where a person's worldview, time axis, and view of humanity all come out. So, if you let AI think without a Philosophy DB, only average "currently trendy business ideas" will come out. If you adopt those, you'll be handing over your life to the average. The Philosophy DB is decisively effective here.

The order of doing things

For those who thought, "I wonder if I can do it too," I'll write the first steps. Even if you don't go as far as business strategy, starting from the X post level is valuable enough.

Step 1: Create a page called "Philosophy DB" in Notion. Inside, create three pages: "General Theory," "Individual Databases," and "Update History." If you use the template I'll distribute at the end, this will be over in an instant.

Step 2: Collect materials. For the first week, roughly gather logs from private chat tools and work chat tools. Prioritize interactions with people you're close to. At the same time, record about 30 minutes of audio on "what you're obsessed with" or "what's been on your mind lately."

Step 3: Let Claude read the collected materials and extract from them. Roughly write down the extracted parts while organizing which of the four blocks of the general theory (Background, Characteristics, Core Philosophy, Practical Logic) they fit into. 10 lines each is fine at first.

Step 4: Try entrusting something to Claude. It could be an X post or a reply to an email. Have it write after referring to the Philosophy DB. See if the result "feels like you."

Step 5: If there's any discomfort, verbalize the true nature of that discomfort and add it to the general theory or individual database. Also record it as a pair in the update history. Repeat this.

The first general theory only needs to be 10 lines. If you aim for perfection, you won't be able to write. Start messy and grow it while operating. This is the correct answer.

The front-end AI can be anything

I've been talking about Claude all this time. I also wrote the order of doing things assuming Claude. But to tell the truth, the front-end AI can be anything.

ChatGPT, Gemini, or Claude—use whichever you like. This is my honest feeling.

Because the essence is external memory.

A human who has an external memory called a Philosophy DB refers it to which AI? That's all it is. The front end is replaceable. If a better model comes out next year, I'll switch. If an even better one comes out the year after, I'll switch again. If the foundation is the same, the switching cost is close to zero.

Why? Because what AI cannot touch is "will." AI can list as many candidate questions as it wants and has as many patterns of writing text as it wants. However, it cannot decide "which question to bet on" or "which one to choose based on which judgment criteria." Will can only be activated by humans. And to activate will, you need a basis. That basis is the external memory called the Philosophy DB.

Conversely, humans who don't have external memory depend on the front-end AI. They become like "I can't write without ChatGPT" or "It has to be Claude." This isn't because they're dependent on AI, but because they haven't put their philosophy outside, so it just looks like they're dependent.

In the current AI industry, there's a huge fuss every time a new model comes out. People are overjoyed or saddened by performance benchmark numbers and get excited about who took the top spot. Everyone is chasing the front end.

I'm not very interested. No matter which one comes out, what I do doesn't change. I have it refer to the Philosophy DB and entrust tasks. That's it. Even if the front end changes, if the foundation is the same, the output will be the same "my writing."

When you realize this, you become free from AI-related information consumption. You don't have to chase fine differences in the latest models. You don't have to react to every new feature someone releases. You can spend your time only on deepening your own foundation without being swayed by others' announcements.

From here on, front-end AI will continue to evolve. Performance gaps will narrow in no time, and roughly the same output will come out no matter which one you use. At that time, what will make the difference? External memory. How deeply, structurally, and in a referable form you have put your own philosophy outside. This is where the difference will be made.

It's not "which AI to use." It's "what kind of external memory do you have?" That is the true question.

Philosophy becomes your own the more you write it

One last thing.

Creating a Philosophy DB might look like preparation for entrusting tasks to AI. But actually, it's not.

The very task of verbalizing your philosophy and putting it outside trains your philosophy. Things that were vague before writing take on a contour by writing. You notice contradictions while writing, organize them, and write again. Through this repetition, philosophy becomes deeper.

Being able to entrust tasks to AI is a bonus. The core is that you become able to hold your own philosophy clearly.

"Giving AI a philosophy" is technically not doing something to the AI, but the act of verbalizing your own philosophy itself. AI is merely a device that gives you the motivation to put it outside.

Write to entrust. Writing trains your philosophy. With a trained philosophy, you can entrust more deeply. If you entrust, you'll want to write again.

In the AI era, only those who turn their philosophy into external memory will win. This is not a hype phrase. It will actually become so in the next few years.

I'm giving away the structure template of the Philosophy DB I use to those who quote-post or repost. Please follow me and let me know in a reply. It's designed so that when you load it into Claude and Notion, it will guide you through designing your own dedicated Philosophy DB.

It's the first step for your philosophy to become your external memory.