Migration, Birth, Incorporation, Independence—A Half-Year I Lived Through with AI

After a long break, you return to the office. Unprocessed documents are piled up on your desk. The moment you open your computer, unread emails and chat notifications cover the screen.

Have you ever looked at a scene like that and felt your head start to hurt?

In spring 2025, in Tokyo, that was exactly where I was.

1. Edogawa-ku, Tokyo — Timezone: Asia/Tokyo (JST)

Let me introduce myself.

I’m a workflow architect based in Takamatsu, Kagawa, designing the “working structures” that allow humans and AI to collaborate—meetings, notes, decisions, and rules.

I run a small studio called Fragment Practice with my wife and our two sons, ages four and zero.

Time moves quickly. Housework, parenting, and business all get packed into a single timeline, and somehow we keep making it work.

Today, I want to look back on the last half-year, during which all of the following happened:

• my wife returning to her hometown for childbirth support
• our relocation to Takamatsu
• the incorporation of a company
• the birth of our second child
• my independence from company employment

And I want to reflect on it together with one more element:

dialogue with AI.

2. Unease before the return home—and early conversations with AI

It was spring 2025. Edogawa-ku, Tokyo. Timezone: Asia/Tokyo (JST).

I was still working in Tokyo as an employee, and I was quietly overwhelmed by what the next few months would require.

My wife would soon return to her family home on Shodoshima. Our second child would be born. Kindergarten for our first son, work, housework, postpartum care—

“Can this really all keep running?”

That sense of unease did not come as a dramatic crisis. It slowly gathered in my chest.

My wife, who is a licensed psychologist and coach, had already become independent a few years earlier.

• She had her own vision
• She worked directly with clients
• She built her work at her own pace

I watched that from the side and often thought:

“Could I ever work like that too?”

But every time I looked directly at reality—

• the cost of living in the Tokyo area
• my current salary
• the scope of responsibility in my job

—the same sentence always won:

“Not now.”

3. The problem of interrupted sessions

Still, moving without thinking felt dangerous.

So I began talking with ChatGPT in order to untangle questions I could no longer carry in my head.

• How do we balance work and parenting?
• If I were to become independent, what forms are realistic?
• Is life in Takamatsu or Shodoshima actually feasible?

When I typed whatever came to mind, ChatGPT could help organize my vague thoughts to a certain extent.

But eventually I hit a problem.

When one session gets long, and I restart in a different session, it stops feeling like a continuation.

I think many people recognize this feeling.

• It is exhausting to explain the same assumptions every time
• The model cannot reliably refer back to past conversations
• It cannot naturally remember “that thing we talked about before”

AI was useful. But as a “long-term partner,” it still felt like something was missing.

4. The origin of fragment_os — preserving context through YAML

That is when I arrived at a simple idea.

What if I summarized the contents of the dialogue and the underlying assumptions in YAML, like a kind of configuration file?

• my profile
• my work and business situation
• family structure and daily rhythm
• values I wanted to protect
• themes I wanted to keep thinking about

If I compressed those into YAML and passed them into the beginning of a session as a prompt, then even when the session changed, the AI became a little closer to:

“someone who could continue from where we left off.”

I no longer had to re-explain the premises every time. Conversations could continue more naturally while carrying prior context.

This way of thinking— sharing an OS-like set of assumptions before starting a dialogue— later became the conceptual source of what I came to call fragment_os and communication_os.

It was not an advanced technology or a complicated system. It was simply this:

write the context down once, and share it clearly with AI.

That alone changed the quality of the dialogue more dramatically than I expected.

5. A dense half-year begins

Once I began working this way, I started deepening conversations with AI across many themes:

• work
• family
• parenting
• the timing of independence
• possible places to relocate

I realized that correctly organizing assumptions and themes for dialogue improved both my own reflection and my conversations with AI.

And in the end, I made a decision:

“I’ll do it myself.”

From that point on, the days became dense, compressed, and accelerating.

6. Return home, relocation to Takamatsu, incorporation, and birth

As I communicated my future plans at work, my wife and our eldest son returned to her family home on Shodoshima.

Meanwhile, from our Tokyo apartment, I was handling all of the following in parallel:

• searching for housing in Takamatsu
• consulting startup support desks
• preparing incorporation procedures
• arranging movers and logistics
• building the Fragment Practice website (Next.js / TypeScript / Vercel)

Each one was something I had never done before.

At that point, I made a decision:

Do not try to hold everything in your head at once.

• inventory the tasks together with AI
• ask it to break procedures into steps
• use it to translate administrative websites and制度 explanations into language I could actually understand

That way, I could keep my focus on just one thing:

what needs to be done today.

7. Ferries, city halls, and one-stroke administrative days

When the time of birth finally arrived, even more tasks were packed into each day.

• procedures at Takamatsu City Hall
• coordination with hospitals in the city
• high-speed ferry travel from Takamatsu to Shodoshima
• procedures at the town office on Shodoshima

Some days, all of that had to be handled almost in a single continuous line.

On days like those, what matters is not “dumping everything onto AI,” but rather “deciding the sequence together with AI.”

• In what order is this realistically possible?
• Is anything missing?
• What can be prepared in advance?

Even for questions too small or too detailed to ask another person, AI stays with you without complaint until the end.

8. October 2025, Takamatsu — Asia/Tokyo (JST)

And then came October 2025. Takamatsu, Kagawa. Timezone: Asia/Tokyo (JST).

I had become independent, operating a small studio called Fragment Practice, and I was now proposing practical ways for humans and AI to work together.

I do not think I arrived here through a beautiful abstract story.

I think the real reason is much simpler:

Through dialogue with AI, I kept deciding for myself who I am and what I can do.

9. Everything starts by writing down assumptions

If your current environment is not in order, if daily tasks alone already consume all your energy, if you want to talk with AI but do not know how to begin—

you do not need to start by writing the “correct” prompt.

Start with something smaller:

• What exactly am I struggling with?
• What assumptions and constraints am I carrying?
• What would make things even slightly easier?

Write those down, even in a sentence each. Then hand them to AI.

Putting assumptions into words and sharing them helps you organize yourself, and also raises the resolution of AI’s answers.

Over this half-year, I felt that again and again.

10. A small proposal from Fragment Practice

If you want to deepen the density of both life and work while collaborating with AI as a partner, then perhaps the OS and dialogue templates we use at Fragment Practice can serve as one reference point.

It does not need to be anything grand.

You can begin with a very small YAML file that contains only:

• your current situation
• what you want to think through next
• what you want AI to help with

A half-year that changes the direction of someone’s life can begin from something that small.

And whether the name of that partner is “Chat-kun,” or something else entirely, probably does not matter very much.