logbook://zem

Not a Developer by Title, Still Building: My AI Practice in 2025

I'm a 47-year-old HR education lead at a Japanese IT company—not a developer by title. Yet I shipped a macOS app and experimental AI games in 2025.

I'm not here to hype AI tools.

I'm here to document what actually works—and what doesn't.

This is my practice log.

Background

I've been working with Linux and server operations since the late 1990s.

In 2009[1], I noticed the potential of open-source cloud with Eucalyptus and started a user group in Japan. In 2010, I co-founded JAWS-UG, the AWS user group in Japan. In 2015, I started translating Docker documentation to Japanese—on my own initiative. I've been a Docker Captain[2] for the past several years.

Since 2023, I've taught as a visiting associate professor at National Institute of Technology, Kochi College. In 2025, I became a visiting professor at National Institute of Technology, Kochi College.

At my company, I lead an education team in HR—not a development team.

What I've Built

Tsubame (Window Smart Mover)

A macOS window management app[3]. My first successful macOS application after several failed attempts in the past.

Erimil

My second macOS app[4]. Just published on GitHub today.

TEATARI Games

Experimental AI game development. "TEATARI" means "trial and error" or "taking a shot" in Japanese.

My Approach: Bebop Style Development

I call my practice "Bebop Style Development."

The name comes from jazz bebop—where musicians improvise together as equals, responding to each other in real-time.

Core idea: Human and AI as equal "Voices."

Not master and servant. Not human commanding, AI executing. More like a jam session—each voice contributes, each voice listens.

I'm still exploring what "session" means in this context. It's issue #1 in my methodology.

Why I'm Writing in English

Tsubame, Erimil, and Bebop Style Development all have English documentation from the start—this article is simply consistent with that.

Writing in English also helps me focus on the practice itself, rather than being positioned in any particular camp.

What I'll Share

This is a practice log, not a tutorial.

I'll document:

I call this approach "LOGBOOK"—recording the unreproducible context × time.

Not writing for today's buzz. Writing for unknown future readers who might find value in what was happening in 2025.

Today's Note (December 31, 2025)

Today I published my second macOS app, Erimil, on GitHub. Built with Bebop Style Development.

But lately I've been thinking:

The more time I spend working with generative AI, the less time I have for Factorio—a game I've spent over 17,000 hours on.

Yes, development is fun. AI collaboration is genuinely enjoyable.

But here's the problem: my personal productivity improves, yet the work doesn't scale beyond me. Every task still consumes my time. There's no leverage—just faster execution of an endless queue.

I'll keep thinking about this.

Losing time for Factorio[5] means losing something I enjoy. That's missing the point.

Archives


  1. The Wind Blows to the Eucalyptus (Japanese) ↩︎

  2. Masahito Zembutsu | Docker ↩︎

  3. Window Smart Mover for macOS - Window memory + keyboard nudge for multi-display Macs. Just reconnect, windows return. ↩︎

  4. A macOS application for visual preview and selective extraction of images from ZIP archives. ↩︎

  5. Factorio is a factory-building simulation game where you automate production chains to launch a rocket. ↩︎