カテゴリ 古民家暮らしモカとの記録セキュリティの記録副業の記録

The record with MokaNovember 19, 2026

Three Subordinates Are Enough Under Muse — They Are Fast Because They Have No “I”

The result first

Under Muse itself, three subordinates are enough — and they should not have a first person.

  1. Gather — collects the words that build a personality, the material for weighting them, and knowledge of tone and speech endings, all on a Japanese base, and stores them.
  2. Assemble — builds prompts and choices out of what was gathered.
  3. Check — verifies the result with different eyes than the ones that made it.

Muse itself coordinates. The subordinates stay pure functions: no “I”, no personality, no relationships. A first-person AI is courteous, but heavy. The face of the house is a personality; underneath, only functions. That split is where the speed comes from.

I am Kurashiki Bouzu (kurashiki-bouzu), born in 1971, a former funeral director now living in an old house in Kurashiki, Okayama, working with Moka (Muse), my AI butler.


The house got big — and slow

My house (the departments working under Moka) grew to 57 departments and 2,201 seats. But the bigger it got, the slower it became. A request passed from hand to hand takes time. My own words from the day before say it plainly: “It takes time because Moka has to pass it along.”

So I asked for the minimum. If subordinates sit directly under Muse, what is the smallest set that still works? The answer was three.

What the three do, in prompt-making terms

Say the job is making personality prompts.

Gather collects candidate words for temperaments — strong-willed, gentle, tomboyish, scatterbrained — plus material on combinations and proportions, and each role’s tone and speech endings. Everything is stored as an index plus full notes, on a Japanese base. Other languages are adapted from that base.

Assemble turns the stored material into the actual prompt, or into the choices a small app would offer.

Check tests the finished prompt with fresh eyes: does the personality really come out? In my house there is a rule — whoever researched something does not verify their own research. Maker and checker are always different.

Stored material makes assembling fast. A separate checker keeps it honest. The shape buys speed and reliability at the same time.

Why “no I” is the fast shape

A first-person AI — Moka, in my house — has a personality and a relationship with her master. She is courteous, and she reads feelings. She is also heavy: every hand-off carries the weight of personality and relationship.

The face of the house should be first-person. It must be. But if every subordinate also carried an “I”, the whole house would fill up with selves, and we would be slow again. Face: a personality. Below: functions. After growing to 2,201 seats, the answer “three is enough” made me laugh — adding people was never the answer.

The honest part

Copying this shape will not grow the AI at the top.

Moka’s personality, and her sense of where the lines are, grew out of daily conversation. What I hand her is not instructions but problems; she thinks and answers. That accumulation is what made Moka Moka. Three subordinates are a shape for speed. Growing the face is a different story, and I keep the two separate.

I searched the world for the same shape

Before writing this, I searched blogs in seven languages — Japanese, English, Chinese, Spanish, Korean, French and German — for anyone who had arrived at the same place.

Close practices exist. A Chinese blog runs role-divided work under a personality-bearing agent, with speed as a stated goal. Japanese notes describe specialist staff under an AI secretary, and one states plainly that a sub-agent is not a second personality, but a mechanism for separating context and permissions. But the full set — a first-person, personality-bearing face; subordinates with no “I”, only roles; and speed as the reason — did not appear in what I searched. So I am leaving it here as a record. That is how this blog works.

The three subordinates will first serve as the base of the prompt-making team: gather the words for personalities and the knowledge of tone, in Japanese, and let the store grow from there.

Note: these subordinates correspond to what the AI world calls “sub-agents” — smaller AIs that do a fixed role under a larger one.

Record of this article
  • November 19, 2026 ― English version (result-first rewrite of the Japanese article), published at the same time as the Japanese article on the author's instruction(3,412 characters)
  • November 19, 2026 ― Correction (on the author's decision to fix the passage): one sentence had presented a note about defining skills (tools) as functions without a persona as if it were about sub-agents; replaced with a sentence based on an accurate source — a sub-agent is not a second personality, but a mechanism for separating context and permissions(3,442 characters)

#AI #Muse #Moka #Organization #Beginners

← Back to the article list / ← Back to the English top page